Mathematics of the Rubik's cube

May 22, 1992 - is called the n × n identity matrix and denoted I or In. ...... [HR] G. Hardy and S. Ramanujan, ”Asymptotic formulae in combinatory analysis”, Proc.
810KB taille 1 téléchargements 370 vues
Mathematics of the Rubik’s cube Associate Professor W. D. Joyner Spring Semester, 1996-7

2 Abstract These notes cover enough group theory and graph theory to understand the mathematical description of the Rubik’s cube and several other related puzzles. They follow a course taught at the USNA during the Fall and Spring 1996-7 semesters.

”By and large it is uniformly true that in mathematics that there is a time lapse between a mathematical discovery and the moment it becomes useful; and that this lapse can be anything from 30 to 100 years, in some cases even more; and that the whole system seems to function without any direction, without any reference to usefulness, and without any desire to do things which are useful.” John von Neumann COLLECTED WORKS, VI, p489

For more mathematical quotes, see the first page of each chapter below, [M], [S] or the www page at http://math.furman.edu/~mwoodard/mquot.html

3 ”There are some things which cannot be learned quickly, and time, which is all we have, must be paid heavily for their acquiring. They are the very simplest things, and because it takes a man’s life to know them the little that each man gets from life is costly and the only heritage he has to leave.” Ernest Hemingway

4

Contents 0 Introduction

11

1 Logic and sets 13 1.1 Logic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 1.1.1 Expressing an everyday sentence symbolically . . . . . 16 1.2 Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 2 Functions, matrices, relations and counting 2.1 Functions . . . . . . . . . . . . . . . . . . . 2.2 Functions on vectors . . . . . . . . . . . . . 2.2.1 History . . . . . . . . . . . . . . . . . 2.2.2 3 × 3 matrices . . . . . . . . . . . . . 2.2.3 Matrix multiplication, inverses . . . . 2.2.4 Muliplication and inverses . . . . . . 2.3 Relations . . . . . . . . . . . . . . . . . . . . 2.4 Counting . . . . . . . . . . . . . . . . . . . .

. . . . . . . .

19 19 24 24 26 26 27 27 31

3 Permutations 3.1 Inverses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 Cycle notation . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3 An algorithm to list all the permutations . . . . . . . . . . . .

33 36 41 46

4 Permutation Puzzles 4.1 15 puzzle . . . . . . . . . . . 4.2 Devil’s circles (or Hungarian 4.3 Equator puzzle . . . . . . . 4.4 Rainbow Masterball . . . . . 4.5 2 × 2 Rubik’s cube . . . . .

51 52 55 57 60 65

. . . . rings) . . . . . . . . . . . . 5

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . . . . .

. . . . .

. . . . . . . .

. . . . .

. . . . . . . .

. . . . .

. . . . . . . .

. . . . .

. . . . . . . .

. . . . .

. . . . . . . .

. . . . .

. . . . . . . .

. . . . .

. . . . . . . .

. . . . .

. . . . . . . .

. . . . .

. . . . .

6

CONTENTS 4.6

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

69 70 71 74 77 77 81 86

5 Groups, I 5.1 The symmetric group . . . . . . . 5.2 General definitions . . . . . . . . 5.2.1 The Gordon game . . . . . 5.3 Subgroups . . . . . . . . . . . . . 5.4 Example: The dihedral group . . 5.5 Example: The two squares group 5.6 Commutators . . . . . . . . . . . 5.7 Conjugation . . . . . . . . . . . . 5.8 Orbits and actions . . . . . . . . 5.9 Cosets . . . . . . . . . . . . . . . 5.10 Dimino’s Algorithm . . . . . . . . 5.11 Permutations and campanology .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

89 90 91 96 98 100 101 104 106 108 112 114 116

. . . . . .

123 . 125 . 128 . 129 . 130 . 132 . 138

. . . . . .

139 . 139 . 141 . 144 . 145 . 148 . 151

4.7 4.8 4.9 4.10 4.11 4.12

3 × 3 Rubik’s cube . . . . 4.6.1 The superflip game 4 × 4 Rubik’s cube . . . . Skewb . . . . . . . . . . . n × n Rubik’s cube . . . . Pyraminx . . . . . . . . . Megaminx . . . . . . . . . Other permutation puzzles

. . . . . . . .

. . . . . . . .

. . . . . . . .

6 Graphs and ”God’s algorithm” 6.1 Cayley graphs . . . . . . . . . . . . . . . . 6.2 God’s algorithm . . . . . . . . . . . . . . . 6.2.1 History . . . . . . . . . . . . . . . . 6.3 The graph of the 15 puzzle . . . . . . . . . 6.3.1 General definitions . . . . . . . . . 6.4 Remarks on applications, NP-completeness

. . . . . .

. . . . . .

. . . . . .

. . . . . .

7 Symmetry groups of the Platonic solids 7.1 Descriptions . . . . . . . . . . . . . . . . . . . . . 7.2 Background on symmetries in 3-space . . . . . . . 7.3 Symmetries of the tetrahedron . . . . . . . . . . . 7.4 Symmetries of the cube . . . . . . . . . . . . . . . 7.5 Symmetries of the dodecahedron . . . . . . . . . . 7.6 Appendix: Symmetries of the icosahedron and S6

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

CONTENTS

7

8 Groups, II 8.1 Homomorphisms . . . . . . . . . . . . . . . . . . . 8.2 Homomorphisms arising from group actions . . . . 8.3 Examples of isomorphisms . . . . . . . . . . . . . . 8.3.1 Conjugation in Sn . . . . . . . . . . . . . . . 8.3.2 Aside: Automorphisms of Sn . . . . . . . . . 8.4 Kernels and normal subgroups . . . . . . . . . . . . 8.5 Quotient subgroups . . . . . . . . . . . . . . . . . . 8.6 Direct products . . . . . . . . . . . . . . . . . . . . 8.6.1 First fundamental theorem of cube theory . 8.6.2 The twists and flips of the Rubik’s cube . . 8.6.3 The slice group of the Rubik’s cube . . . . . 8.7 Semi-direct products . . . . . . . . . . . . . . . . . 8.8 Wreath products . . . . . . . . . . . . . . . . . . . 8.8.1 Application to order of elements in Cm wr Sn 9 The Rubik’s cube and the word problem 9.1 Background on free groups . . . . . . . . . . . 9.1.1 Length . . . . . . . . . . . . . . . . . . 9.1.2 Trees . . . . . . . . . . . . . . . . . . . 9.2 The word problem . . . . . . . . . . . . . . . 9.3 Generators and relations . . . . . . . . . . . . 9.4 Generators, relations for groups of order < 26 9.5 The presentation problem . . . . . . . . . . . n 9.6 A presentation for Cm >¢ Sn+1 . . . . . . . . 9.6.1 A proof . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

153 . 153 . 156 . 157 . 159 . 160 . 161 . 163 . 165 . 166 . 167 . 168 . 173 . 176 . 178

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

181 . 181 . 182 . 183 . 184 . 185 . 187 . 192 . 193 . 195

10 The 2 × 2 and 3 × 3 cube groups 10.1 Mathematical description of the 3 × 3 cube moves 10.1.1 Notation . . . . . . . . . . . . . . . . . . . 10.1.2 Corner orientations . . . . . . . . . . . . . 10.1.3 Edge orientations . . . . . . . . . . . . . . 10.1.4 The semi-direct product . . . . . . . . . . 10.2 Second fundamental theorem of cube theory . . . 10.2.1 Some consequences . . . . . . . . . . . . . 10.3 Rubika esoterica . . . . . . . . . . . . . . . . . . . 10.3.1 Coxeter groups . . . . . . . . . . . . . . . 10.3.2 The moves of order 2 . . . . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . .

197 197 197 200 201 202 203 207 208 210 211

8

CONTENTS 10.4 Mathematical description of the 2 × 2 cube moves . . . . . . . 212

11 Other Rubik-like puzzle groups 11.1 On the group structure of the skewb . . . 11.2 On the group structure of the pyraminx . 11.2.1 Orientations . . . . . . . . . . . . . 11.2.2 Center pieces . . . . . . . . . . . . 11.2.3 The group structure . . . . . . . . 11.3 A uniform approach . . . . . . . . . . . . . 11.3.1 General remarks . . . . . . . . . . 11.3.2 Parity conditions . . . . . . . . . . 11.4 The homology group of the square 1 puzzle 11.4.1 The main result . . . . . . . . . . . 11.4.2 Proof of the theorem . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

215 . 215 . 219 . 221 . 223 . 224 . 225 . 225 . 226 . 227 . 228 . 231

12 Interesting subgroups of the cube group 12.1 The squares subgroup . . . . . . . . . . . . . . . . 12.2 P GL(2, F5 ) and two faces of the cube . . . . . . . 12.2.1 Finite fields . . . . . . . . . . . . . . . . . 12.2.2 M¨obius transformations . . . . . . . . . . 12.2.3 The main isomorphism . . . . . . . . . . . 12.2.4 The labeling . . . . . . . . . . . . . . . . . 12.2.5 Proof of the second theorem . . . . . . . . 12.3 The cross groups . . . . . . . . . . . . . . . . . . 12.3.1 P SL(2, F7 ) and crossing the cube . . . . . 12.3.2 Klein’s 4-group and crossing the pyramnix

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . .

251 . 252 . 252 . 254 . 255 . 256 . 258 . 259 . 259 . 261

13 Crossing the Rubicon 13.1 Doing the Mongean shuffle . . 13.2 Background on P SL2 . . . . . 13.3 Galois’ last dream . . . . . . . 13.4 The M12 generation . . . . . . 13.5 Coding the Golay way . . . . 13.6 M12 is crossing the rubicon . . 13.7 An aside: A pair of cute facts 13.7.1 Hadamard matrices . . 13.7.2 5-transitivity . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . . . .

. . . . . . . . .

. . . . . . . . . . .

. . . . . . . . .

. . . . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

233 234 236 236 240 243 244 245 246 246 249

CONTENTS 14 Appendix: Some solution strategies 14.1 The subgroup method . . . . . . . . . . . . 14.1.1 Example: the corner-edge method . . 14.1.2 Example: Thistlethwaite’s method . 14.2 3 × 3 Rubik’s cube . . . . . . . . . . . . . . 14.2.1 Strategy for solving the cube . . . . . 14.2.2 Catalog of 3 × 3 Rubik’s ”supercube” 14.3 4 × 4 Rubik’s cube . . . . . . . . . . . . . . 14.4 Rainbow masterball . . . . . . . . . . . . . . 14.4.1 A catalog of rainbow moves . . . . . 14.5 Equator puzzle . . . . . . . . . . . . . . . . 14.6 The skewb . . . . . . . . . . . . . . . . . . . 14.6.1 Strategy . . . . . . . . . . . . . . . . 14.6.2 A catalog of skewb moves . . . . . . 14.7 The pyraminx . . . . . . . . . . . . . . . . . 14.8 The megaminx . . . . . . . . . . . . . . . . 14.8.1 Catalog of moves . . . . . . . . . . .

9

. . . . . . . . . . . . . . . . . . . . moves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

263 . 263 . 264 . 265 . 266 . 266 . 267 . 267 . 269 . 270 . 271 . 273 . 273 . 273 . 274 . 275 . 276

10

CONTENTS

Chapter 0 Introduction ”The advantage is that mathematics is a field in which one’s blunders tend to show very clearly and can be corrected or erased with a stroke of the pencil. It is a field which has often been compared with chess, but differs from the latter in that it is only one’s best moments that count and not one’s worst.” Norbert Wiener EX-PRODIGY: MY CHILDHOOD AND YOUTH

Groups measure symmetry. No where is this more evident than in the study of symmetry in 2- and 3-dimensional geometric figures. Symmetry, and hence groups, play a key role in the study of crystallography, elementary particle physics, coding theory, and the Rubik’ s cube, to name just a few. This is a book biased towards group theory not the ”the cube”. To paraphrase the German mathematician David Hilbert, the art of doing group theory is to pick a good example to learn from. The Rubik’s cube will be our example. We motivate the study of groups by creating a group-theoretical model of Rubik’s cube-like puzzles. Although some solution strategies are discussed (for the Rubik’s cube - the 3 × 3 × 3 and 4 × 4 × 4 versions, the ”Rubik tetrahedron” or pyraminx, the the ”Rubik dodecahedron” or megaminx, the skewb, square 1, the masterball, and the equator puzzle), these are viewed more abstractly than most other books on the subject. We regard a solution strategy merely as a not-too-inefficient algorithm for producing all the elements in the associated group of moves. 11

12

CHAPTER 0. INTRODUCTION

The approach here is different than some other group theory presentations (such as Rotman [R]) in that (a) we stress puzzle-related examples over general theory, (b) we present some of the basic notions algorithmically (as in [Bu]), and (c) we tried to keep the level as low as possible for as long as possible (though only the reader can judge if we’ve keep it low enough long enough or too low too long). The hope is that by following along and doing as many of the exercises as possible the reader will have fun learning about how groups occur in this area of recreational mathematics. Along the way, we shall also explain the rules of a two person game (the Gordon game) which arises from group theory a game which arises from a group not a group which arise from a game. See also the superflip game introduced in §4.6 and the Icosian game in §6.2.1. Acknowledgements: Many of the graphs given below were produced with the help of MAPLE, LVIEW, and Windows 95 PBRUSH. Some of the grouptheoretical calculations (such as the order of a group element) were determined with the help of GAP. The comments from my students were a big help. I especially thank Ann Luers and Jim McShea for their help and collaboration on some of the topics. I’d also like to thank Andy Southern for collaborating on some of the masterball material [JS] which appears in chapter 14 and Dennis Spellman for the argument in §9.7 where a presentation of wreath products is derived.

Chapter 1 Logic and sets ”If logic is the hygiene of the mathematician, it is not his source of food; the great problems furnish the daily bread on which he thrives.” Andre Weil ”The future of mathematics”, AMER MATH MONTHLY, May, 1950 ”The rules of logic are to mathematics what those of structure are to architecture.” Bertrand Russell MYSTICISM AND LOGIC AND OTHER ESSAYS, 1917, p61 This chapter will present some background to make some of the terminology and notation introduced later a little clearer. It is not intended to be a rigorous introduction to mathematical logic.

1.1

Logic

A statement is a logical assertion which is either true or false. (Of course we assume that this admitedly circular ’definition’ is a statement.) Sometimes the truth or falsity of a statement is called its Boolean value. One can combine several statements into a single statement using the connectives ’and’, ’or’, and ’implies’. The Boolean value of a statement is changed using the ’negation’. We shall also use ’if and only if’ and ’exclusive or’ but these can 13

14

CHAPTER 1. LOGIC AND SETS

be defined in terms of negation ∼ and the other three connectives (∨, ∧, and ⇒). Exercise 1.1.1 Do this. In other words, express − ∨ and ⇐⇒ in terms of ∼, ∨, ∧, and ⇒. Notation: Let p and q be statements. statement

notation

terminology

p and q p or q p implies q negate p p if and only if q either p or q (not both)

p∧q p∨q p⇒q ∼p p ⇐⇒ q p −∨ q

”conjunction” ”disjunction” ”conditional” ”negation” ”if and only if” ”exclusive or”

Truth tables: Given the Boolean values of p, q, we can determine the values of p ∧ q, p ∨ q, p ⇒ q, p ⇐⇒ q, p− ∧ q using the truth tables:

p

p

q

T T F F

T F T F q

T T T F F T F F p

p∧q T F F F p⇒q

T T T F F T F F

q

p∨q

T T T T F T F T T F F F p q p ⇐⇒ q

T F T T q

p

T T T F F T F F

T F F T

p −∨ q F T T F

p

∼p

T F

F T

1.1. LOGIC

15

Exercise 1.1.2 (M. Gardner) Determine which of the following statements is true. • Exactly one of these statements is false. • Exactly two of these statements are false. • Exactly three of these statements are false. • Exactly four of these statements are false. • Exactly five of these statements are false. • Exactly six of these statements are false. • Exactly seven of these statements are false. • Exactly eight of these statements are false. • Exactly nine of these statements are false. • Exactly ten of these statements are false. Definition 1 Let B be the set {0, 1} with the two operations addition (+) and multipication (*) defined by the following tables p

q

p+q

p

q

p*q

1 1 0 0

1 0 1 0

0 1 1 0

1 1 0 0

1 0 1 0

1 0 0 0

(Note how these mimic the truth tables of ’exclusive or’ (− ∨) and ’and’ (∧).) We call B the Boolean algebra. Exercise 1.1.3 Use truth tables to verify DeMorgan’s laws: (a) p ∧ (q ∨ r) ⇐⇒ (p ∧ q) ∨ (p ∧ r) , (b) p ∨ (q ∧ r) ⇐⇒ (p ∨ q) ∧ (p ∨ r), and the laws of negation: (c) ∼ (p ∧ q) ⇐⇒ (∼ p) ∨ (∼ q) , (d) ∼ (p ∨ q) ⇐⇒ (∼ p) ∧ (∼ q). (You may want to do (a), (c), (d) first, then deduce (b) from these.)

16

CHAPTER 1. LOGIC AND SETS

Definition 2 A logical argument is a sequence of statements (called hypotheses) p1 , p2 , ..., pn , which imply a statement q (called the conclusion). In other words, a logical argument is a true statement of the form (p1 ∧ p2 ∧ ... ∧ pn ) ⇒ q. Exercise 1.1.4 Use truth tables to verify the logical argument ((p ⇒ q) ∧ (q ⇒ r)) ⇒ (p ⇒ r). Definition 3 ‘For all‘, written ∀, is the universal quantifier. ‘There exists‘, written ∃, is the existential quantifier. A variable is a letter denoting some (possibly unknown) object. A constant is a letter denoting some specific, well-defined object. A term is a variable or constant. A predicate is a function P : {terms} → {logical statements}. Example 4 ‘Daffy Duck‘ is a constant. ‘x and Daffy Duck are cartoon characters‘ is a predicate involving a variable and a constant.

1.1.1

Expressing an everyday sentence symbolically

When creating a model of the Rubik’s cube, we shall of course need to convert some ‘everyday statements‘ into symbolical form in order to perform mathematical analysis. Let’s illustrate this with an example. Example 5 Consider the statement ”Each student in this class is a mathematics major”. Let M (x) = x is a mathematics major,

S(x) = x is a student in this class.

The symbolic form is ∀x, S(x) ⇒ M (x). Exercise 1.1.5 Convert ”Someone in this class likes the Rubik’s cube” to symbolic form.

1.2. SETS

17

Exercise 1.1.6 (M. Gardner [Gar2] Professor White, Professor Brown and Professor Black were lunching together. ”Isn’t it remarkable”, said the lady, ”that our names are white black and brown and one of us has black hair, one has brown hair, and one has white hair.” ”It is indeed”, said the one with the black hair, ”and have you noticed that not one has hair color to match our name?” The lady’s hair is not brown. What is the color of Professor Black’s hair?

1.2

Sets

A set is a ’well-defined’ collection of objects. The objects in a set are the elements of the set. There are two common ways to describe a set: (a) list all its elements (if the set is finite, e.g., {1, 2, 3}), (b) describe the set using properties of its elements (e.g., the set of all even integers can be described by {n | n an integer, 2|n}). Remark 1 We must be a little careful when describing sets using properties since some ‘self-referential‘ properties lead to contraditions: let R = {x | x ∈ / x}. In other words, for all x, x ∈ R ⇐⇒ x ∈ / x. In particular, if we take x = R then this becomes R ∈ R ⇐⇒ R ∈ / R, an obvious contradition. The problem is that R is not ”well-defined” (in the sense that it does not satisfy the set-theoretic axioms which we will skip here). The empty set is the set containing no elements, denoted ∅. Notation: Let S and T be sets. Assume that S ⊂ X. statement

notation

terminology

set of elements in S and T set of elements in S or T set of elements in S or T (not both) set of elements not in S S is a subset of T

S∩T S∪T S∆T Sc S⊂T

intersection union symmetric difference complement subset

18

CHAPTER 1. LOGIC AND SETS

Exercise 1.2.1 Use Venn diagrams to verify the DeMorgan laws: (a) S ∩ (T ∪ U ) = (S ∩ T ) ∪ (S ∩ U ) , (b) S ∪ (T ∩ U ) = (S ∪ T ) ∩ (S ∪ U ) , and the laws of negation: (c) (S ∩ T )c = S c ∪ T c , (d) (S ∪ T )c = S c ∩ T c . Definition 6 We call two sets S, T disjoint if they have no elements in common, i.e., if S ∩ T = ∅. If S = ∪ni=1 Si , where the S1 , S2 , ..., Sn are pairwise disjoint sets then we call this union a partition of S. Example 7 If S = Z, S1 = {..., −2, 0, 2, ...} = even integers, S2 = {..., −3, −1, 1, 3, ...} = odd integers, then S = S1 ∪ S2 is a partition of the integers into the set of even and odd ones.

Logic/set theory analogs: Just as one can use connectives to form new statements from old statements, there are analogous ways to form new sets from old ones using ’intersection’ (the analog of ’and’), ’union’ (the analog of ’or’), and ’complement’ (the analog of ’negation’). The analog of ’implies’ is ’subset’. set theory sets union intersection subset symmetric difference equal Venn diagrams

logic statements or and implies exclusive or if and only if truth tables

For more on logic or set theory, see for example [C] or [St].

Chapter 2 Functions, matrices, relations and counting ”I think mathematics is a vast territory. The outskirts of mathematics are the outskirts of mathematical civilization. There are certain subjects that people learn about and gather together. Then there is a sort of inevitable development in those fields. You get to a point where a certain theorem is bound to be proved, independent of any particular individual, because it is just in the path of development.” William P. Thurston MORE MATHEMATICAL PEOPLE, NY, 1990, p332 ”Chance favors the prepared mind.” Louis Pasteur

This chapter will introduce some frequently used notions.

2.1

Functions

Let S and T be finite sets. Definition 8 : A function f from S to T is a rule which associates to each element s ∈ S exactly one element t ∈ T . We will use the following notations 19

20CHAPTER 2. FUNCTIONS, MATRICES, RELATIONS AND COUNTING for this: f :S→T (”f is a function from S to T”), f : s 7−→ t (”f sends s in S to t in T”), t = f (s) (”t is the image of s under f”). A function is also called a map, mapping, or transformation. We call S the domain of f , T the range of f , and the set f (S) = {f (s) ∈ T | s ∈ S} the image of f. The Venn diagram depicting this setup is:

Venn diagram of f : S → T The Cartesian product of two sets S, T is the set of pairs of elements taken from these sets: S × T = {(s, t) | s ∈ S, t ∈ T }. Example 9 If R denotes the set of all real numbers then the Cartesian product R × R is simply the plane of pairs of real numbers we are all familiar with.

2.1. FUNCTIONS

21

More generally, we may iterate this process and take the Cartesian product the set of real numbers with itself n times (where n > 1 is any integer) to get R × ... × R (n times). This n-fold Cartesian product is denoted more conveniently by Rn . An element of the set Rn will be called a vector or an n-vector to be specific. The graph of a function f : S → T is the subset {(s, f (s) | s ∈ S} ⊂ S × T. It is not possible for every subset of S × T the graph of some function from S to T . The following fact classifies exactly which subsets of S × T can arise as the graph of a function from S to T . Lemma 10 : Let X ⊂ S × T . X is the graph of a function from S to T if and only if, for all (s1 , t1 ) ∈ X and (s2 , t2 ) ∈ X, whenever t1 6= t2 we also have s1 6= s2 . Exercise 2.1.1 Verify this. (Hint: Let pr1 : S × T → S (s, t) 7−→ s be projection onto the 1st component. Recall that the graph of a function has the property that pr1−1 (s) is always a singleton.) Definition 11 If the image of the function f : S → T is all of T , i.e., if f (S) = T , then we call f surjective (or ”onto”, or ”is a surjection”). Equivalently, a function f from S to T is surjective if each t in T is the image of some s in S under f. Occasionally, you see the following special notation for surjective functions: f :S→ À T

(f ”maps S surjectively onto T ”).

Exercise 2.1.2 State a general rule which determines those subsets X of S × T which are the graph of some surjective function from S to T . (Use the projection onto the second component, pr2 : S × T → T (s, t) −→ t in your rule.)

22CHAPTER 2. FUNCTIONS, MATRICES, RELATIONS AND COUNTING Question: Suppose that |S| < |T |. Is there a surjective function f : S → T ? Explain. Definition 12 : A function f : S → T is called injective (or ”one-to-one” or ”an injection”) if each element t belonging to the image f (S) is the image of exactly one s in S. In other words, f is an injection if the condition f (s1 ) = f (s2 ) (for some s1 , s2 ∈ S) always forces s1 = s2 . Sometimes the ”hook arrow” notation is used to denote an injective function f : S ,→ T. Question: Suppose that |S| > |T |. Is there an injective function f : S → T ? Explain. Exercise 2.1.3 Suppose that |S| = |T |. Show that a function f : S → T is surjective if and only if it is injective. Definition 13 A function f : S → T is called a bijection if it is both injective and surjective. Equivalently, a bijection from S to T is a function for which each t in T is the image of exactly one s in S. Exercise 2.1.4 Give an algorithm for determining if a given finite subset X of S × T is the graph of a bijection from S to T . Definition 14 A set S is called countable if there exists a bijection f : S → Z to the set of integers Z. Example 15 The set S of all rational numbers 0 < r < 1 is countable since you can define f : S → Z as follows: f (1) = 1/2, f (2) = 1/3, f (3) = 2/3, f (4) = 1/4, f (5) = 3/4, f (6) = 1/5, f (7) = 2/5, f (8) = 3/5, and so on. There are φ(n) terms of the form m/n, where m is relatively prime to n and φ(n) denotes the number of positive integers less than or equal to n which are relatively prime to n (i.e., have no common prime divisors). (φ is sometimes called Euler’s phi function).

2.1. FUNCTIONS

23

Exercise 2.1.5 Give an algorithm for determining if a given finite subset X of S × T is the graph of a bijection from S to T . Example 16 Let C be the cube in 3-space having vertices at the points O = (0, 0, 0), P1 = (1, 0, 0), P2 = (0, 1, 0), P3 = (0, 0, 1), P4 = (1, 1, 0), P5 = (1, 0, 1), P6 = (0, 1, 1), P7 = (1, 1, 1). We shall also (to use a notation which will be used more later) denote these by dbl, dfl, dbr, ubl, dfr, ufl, ubr, ufr, resp. Let C0 = {O, P1 , ..., P7 } be the set of the 8 vertices of C, let C1 = {uf, ur, ub, ul, f r, br, bl, f l, df, dr, db, dl} be the set of the 12 edges of C, and let C2 = {F, B, U, B, L, R} be the set of the 6 faces of C. Let r be the rotation of a point (x, y, z) by 180 degrees about the passing through the points (1/2, 1/2, 0), (1/2, 1/2, 1). Note that r : R3 → R3 is a function which sends the cube C onto itself. The cube C is pictured as follows:

cube in 3-space This function r induces three functions f0 : C0 → C0 ,

f1 : C1 → C1 ,

f2 : C2 → C2 .

24CHAPTER 2. FUNCTIONS, MATRICES, RELATIONS AND COUNTING where fi is the function which sends x ∈ Ci to its image r(x) under r (which is again in Ci ), for i = 0, 1, 2. Each fi is a bijection. Exercise 2.1.6 Finish the labeling of the vertices and the faces in the above picture and describe f1 and f2 explicitly by filling out the tables v

f0 (v) e

f1 (e) f

f2 (f )

Lemma 17 If f : S → T is a bijection then there exists a function f −1 : T → S defined by the following property: for s ∈ S and t ∈ T , we have f −1 (t) = s

if and only if f (s) = t.

Exercise 2.1.7 Prove this. Definition 18 The function f −1 in the above lemma is called the inverse function of f .

2.2

Functions on vectors

This section presents a few basic facts about matrices, which we regard as a certain type of function which sends vectors to vectors. For further details, see any text on linear algebra, for example [JN] where the historical introduction below is borrowed from.

2.2.1

History

The mathematician who first published a major work which seriously studied matrices and matrix algebra in the western world was Lord Arthur Cayley

2.2. FUNCTIONS ON VECTORS

25

(1821-1895) of Cambridge, England. He wrote a memoir on the theory of linear transformations which was published in 1858 and is often thought of as one of the “fathers” of matrix theory, though in fact it was his friend and colleague James Sylvester who first coined the term “matrix”. Here is one of the earliest examples which motivated Cayley: if we have three coordinate systems (x, y), (x0 , y 0 ), and (x00 , y 00 ) , connected by (

x0 = x + y y0 = x − y

and (

x00 = −x0 − y 0 0 y 0 = −x0 + y 0

then the relationship between (x, y) and (x00 , y 00 ) is given (

x00 = −x0 − y 0 = −(x + y) − (x − y) = −2x 0 y 0 = −x0 + y 0 = −(x + y) + (x − y) = −2y

If we do as Cayley did and abbreviate the three change of coordinates by the square array of the coefficients then we obtain the three arrays "

1 1 1 −1

# "

,

−1 −1 −1 1

# "

,

−2 0 0 −2

#

.

Cayley noticed that there was an algebraic rule which allows us to determine the third array from the first two, without doing the substitution we did above to relate (x00 , y 00 ) to (x, y). This rule involves the rows of the first array and the columns of the second array. For example, to get the entry in the upper left-hand corner of the third array, Cayley explained we need tocombine the first row of the first array with the first column of the second array as follows: −2 = 1 · (−1) + 1 · (−1). (The general formula will be given below.) To get the entry in the upper right-hand corner of the thirdarray, we combine the first row of the first array with the second column of the second array: 0 = 1 · (−1) + 1 · 1. To get the entry in the lower left-hand corner of the third array, we combine the second row of the first array with the first column of the second array: 0 = 1 · (−1) + (−1) · (−1). Finally, to get the entry in the upper left-hand corner of the third array, we combine the first row of the first array with the first column of the second array: −2 = 1 · (−1) + (−1) · 1. The relationship which we just described betweeen the about three arrays he called ”matrix multiplication”.

26CHAPTER 2. FUNCTIONS, MATRICES, RELATIONS AND COUNTING

2.2.2

3 × 3 matrices

First, a 3 × 3 matrix is a 3 × 3 table of real numbers 



a11 a12 a13   A =  a21 a22 a23  a31 a32 a33 which acts on R3 by matrix multiplication: 









a11 a12 a13 x a11 x + a12 y + a13 z      A~v =  a21 a22 a23   y  =  a21 x + a22 y + a23 z  a31 a32 a33 z a31 x + a32 y + a33 z In general, any such 3 × 3 matrix gives rise to a function A : R3 → R3 . Example 19 If A = I3 , the 3 × 3 identity matrix, 



1 0 0    0 1 0  0 0 1 then I3~v = ~v for all ~v ∈ R3 .

2.2.3

Matrix multiplication, inverses

An m × n matrix (of real numbers) is a rectangular array or table of numbers arranged with m rows and n columns. It is usually written:    A=  

a11 a21 .. .

a12 a22

... a1n ... a2n .. .

   .  

am1 am2 ... amn The (i, j)th entry of A is aij . The ith row of A is h

ai1 ai2 · · · ain

i

(1 ≤ i ≤ m)

The jth column of A is      

a1j a2j .. . amj

     

(1 ≤ j ≤ n)

2.3. RELATIONS

27

A matrix having as many rows as it has columns (m = n) is called a square matrix. The square n × n matrix        

1



... 0 ..  .. . 0 .    ..  . 0   .. 0 . 0 1

is called the n × n identity matrix and denoted I or In .

2.2.4

Muliplication and inverses

You can multiply an m × n matrix A by a n × p matrix B and get a m × p matrix AB. The (i, j)th entry of AB is computed as follows: 1. Let k = 1 and c0 = 0. 2. If k = m, you’re done and aij = cm . Otherwise proceed to the next step. 3. Take the k th entry of the ith row of A and multiply it by the k th entry of the j th row of B. Let ck = ck−1 + aik bkj . 4. Increment k by 1 and go to step 2. In other words, multiply each element of row i in A with the corresponding entry of column j in B, add them up, and put the result in the (i, j) position of the array for AB. If A is a square n×n matrix and if there is a matrix B such that AB = In then we call B the inverse matrix of A, denoted A−1 .

2.3

Relations

A relation on a set is a generalization of the concept of a function from S to itself. Definition 20 : Let S be a set. If R is a subset of S × S then we call R a relation on S. If (x, y) ∈ R then we say that x is related to y.

28CHAPTER 2. FUNCTIONS, MATRICES, RELATIONS AND COUNTING We may also regard a relation R on S as a function R : S × S → {0, 1}. In this form, we say x is related to y (for x, y ∈ S) if f (x, y) = 1. This is a very general notion. There are lots and lots of relations in mathematics - inequality symbols, functions, subset symbols are all common examples of relations. Example 21 Let S be any set and let f be a function from S to itself. This function gives rise to the relation R on S defined by the graph of f : R = {(x, y) ∈ S × S | y = f (x), for x ∈ S}. (It is through this correspondence that we may regard a function as a relation.) Example 22 Let S be the set of all subsets of {1, 2, ..., n}. Let R be defined by R = {(S1 , S2 ) | S1 ⊂ S2 , S1 ∈ S, S2 ∈ S}. Note that R is a relation. Exercise 2.3.1 Find a relation corresponding to the symbol < on the real line. Definition 23 Let R be a relation on a set S. We call R an equivalence relation if (a) any element s ∈ S is related to itself (”reflexive”), (b) if s is related to t (i.e., (s, t) belongs to S) then t is related to s (”symmetry”), (c) if s1 is related to s2 and s2 is related to s3 then s1 is related to s3 (”transitivity”). Example 24 The equality symbol = provides an equivalence relation on the real line: let D = {(x, x) | x real}. This is an equivalence relation on the real line: note x = y if and only if (x, y) belongs to D. Notation: If R is an equivalence relation on S then we often write x ∼ y or x ≡ y in place of (x, y) ∈ R, for simplicity.

2.3. RELATIONS

29

Example 25 Fix an integer n > 1. For integers x, y, define x ≡ y if and only if n divides x − y. In this case, we say that x is congruent to y mod n. The equivalence class of x is sometimes (for historical reasons) called the residue class (or congruence class) of x mod n. This notation was first introduced by Gauss 1 Exercise 2.3.2 (a) Let f (x) = x2 , let S be the real line, and let R be the corresponding relation as in the first example. Is R an equivalence relation? (b) Let f (x) = 2x, let S be the real line, and let R be the corresponding relation as in the first example. Is R an equivalence relation? Exercise 2.3.3 Let R be the corresponding relation as in the second example. Is R an equivalence relation? Let R be an equivalence relation on a set S. For s ∈ S, we call the subset [s] = {t ∈ S | s ∼ t} the equivalence class of s in S. Example 26 For integers x, y, define x ≡ y if and only if 3 divides x − y. the equivalence classes are [0] = {..., −6, −3, 0, 3, 6, ...}, [1] = {..., −5, −2, 1, 4, 7, ...}, [2] = {..., −4, −1, 2, 5, 8, ...}. Exercise 2.3.4 Show that for any s1 and s2 in S, we have either (a) [s1 ] = [s2 ], or (b) [s1 ] is disjoint from [s2 ]. As a consequence of this exercise, we see that if R is an equivalence relation on a set S then the equivalence classes of R partition S into disjoint subsets. We state this as a separate lemma for future reference (we also assume S is finite for simplicity): 1

C. F. Gauss, 1777-1855, is regarded by many as one of the top mathematicians of all time. At the age of 21 he wrote ”Disquisitiones Arithmeticae”, which started a new era of number theory and introduced this notation.

30CHAPTER 2. FUNCTIONS, MATRICES, RELATIONS AND COUNTING Lemma 27 : If S is a finite set and R is an equivalence relation on S then there are subsets S1 ⊂ S, S2 ⊂ S, ..., Sk ⊂ S, satisfying the following properties: (1) S is the union of all the Si ’s: S = S1 ∪ S2 ∪ ... ∪ Sk = ∪i Si (2) the Si ’s are disjoint: for 1 ≤ i ≤ k, 1 ≤ j ≤ k, if i 6= j then Si ∩ Sj = ∅. (These last two properties say that the Si ’s partition S in the sense of the previous chapter.) (3) the Si ’s exhaust the collection of equivalence classes of R: for each 1 ≤ i ≤ k, there is an si ∈ S such that Si = [si ]. (This element si is called a representative of the equivalence class Si .) Example 28 For real numbers x, y, define x ≡ y if and only if x − y is an integer. The equivalence classes are of the form [x] = {..., x − 2, x − 1, x, x + 1, x + 2, ...}, for x real. Each equivalence class has exactly one representative in the half open interval [0, 1). Remark 2 Conversely, given a partition as in (1), there is an equivalence relation R such that Si = [si ], for some some si ∈ S, where [s] = {x ∈ S | s ∼ x} is the equivalence class of s with respect to R. Indeed, we define R = ∪ki=1 Si × Si . This is an equivalence relation and s ∼ t if and only if s, t ∈ Si , for some i = 1, 2, ..., k. Exercise 2.3.5 Let S be the set Z of all integers. Let R be the relation defined by (x, y) ∈ R if and only if x − y is an even number (i.e., an integer multiple of 2). (1) Show that this is an equivalence relation, (2) Find the sets Si in the above lemma which partition S, (3) Find a representative of each equivalence class Si .

2.4. COUNTING

2.4

31

Counting

This section quickly surveys the few basic counting principles we shall use later. Addition principle: Let S1 , ..., Sn denote disjoint finite sets. Then |S1 ∪ ... ∪ Sn | = |S1 | + ... + |Sn |. Example 29 If there are n bowls, each containing some distinguishable marbles and if Si is the set of marbles in the ith bowl then the number of ways to pick a marble from exactly one of the bowls is |S1 | + ... + |Sn |, by the addition principle. Corollary 30 (Pigeonhole principle) If there are n objects (pigeons) which must be placed in m (md < n) boxes (pigeonholes) then there is at least one box with at least d + 1 objects. Example 31 If you are in a room with 9 others then there must be either at least 5 people you know or 5 people you don’t know (not counting yourself ). In this case, there are n = 9 objects and m = 2 boxes (the friend box and the stranger box) so we may take d = 4 in the pigeonhole principle. Multiplication principle: Let S1 , ..., Sn denote finite sets. Then |S1 × ... × Sn | = |S1 | · ... · |Sn |. Example 32 If there are n bowls, each containing some distinguishable marbles and if Si is the set of marbles in the ith bowl then the number of ways to pick exactly one marble from each of the bowls is |S1 | · ... · |Sn |, by the multiplication principle. Corollary 33 The number of ordered selections, taken without repetition, of m objects from a set of n objects (m < n) is n! = n · (n − 1) · ... · (n − m + 1). (n − m)! Corollary 34 The number of ordered selections, taken with repetition allowed, of m objects from a set of n objects (m < n) is nm .

32CHAPTER 2. FUNCTIONS, MATRICES, RELATIONS AND COUNTING Example 35 The number of n-tuples, without repetition, of objects from the set {1, 2, ..., n} is n! = n · (n − 1) · ... · 1. Exercise 2.4.1 Let C be a set of 6 distinct colors. Fix a cube in space (imagine it sitting in front of you on a table). We call a coloring of the cube a choice of exactly one color per side. Let S be the set of all colorings of the cube. We say x, y ∈ S are equivalent if x and y agree after a suitable rotation of the cube. (a) Show that this is an equivalence relation. (b) Count the number of equivalence classes in S.

Chapter 3 Permutations ”What we have to learn to do, we learn by doing” Aristotle ETHICS ”Mathematics, springing up from the soil of basic human experience with numbers and data and space and motion, builds up a far-flung architectural structure composed of theorems which reveal insights into the reasons behind appearences and of concepts which relate totally disparate concrete ideas.” Sanders MacLane AMER. MATH. MONTHLY, 1954 Let Tn = {1, 2, ..., n} be the set of integers from 1 to a fixed positive integer n. When n is fixed and there is no ambiguity sometimes we will simply write T for Tn . A permutation of T is a bijection from T to itself. (A bijection was defined in Definition 13.) For example, on the 3 × 3 Rubik’s cube there are 9 · 6 = 54 facets. If you label them 1, 2, ..., 54 (in any way you like) then any move of the Rubik’s cube corresponds to a permutation of T54 . In this chapter we present some basic notation and properties of permutations. Notation: We may denote a permutation f : T → T by a 2 × n array: Ã

f↔

1 2 ... n f (1) f (2) ... f (n) 33

!

34

CHAPTER 3. PERMUTATIONS

Example 36 The identity permutation, denoted by I, is the permutation which doesn’t do anything: Ã

I=

1 2 ... n 1 2 ... n

!

Definition 37 Let ef (i) = #{j > i | f (i) > f (j)},

1 ≤ i ≤ n − 1.

Let swap(f ) = ef (1) + ... + ef (n − 1). We call this the swapping number (or length of the permutation f since it counts the number of times f swaps the inequality in i < j to f (i) > f (j). If we plot a bar-graph of the function f then swap(f ) counts the number of times the bar at i is higher than the bar at j. We call f even if swap(f ) is even and we call f odd otherwise. The number sign(f ) = (−1)swap(f ) is called the sign of the permutation f . Example 38 Let n = 3, so T = {1, 2, 3}. We may describe the permutation f : T → T for which f (1) = 2, f (2) = 1, f (3) by a 2 × 3 array Ã

1 2 3 2 1 3

!

or by a ”crossing diagram”: 1

1 \/ /\

2

2

3 -- 3 The number of crosses in this diagram is the swapping number of f, from which we can see that the permutation is odd.

35 Exercise 3.0.2 Express f : T → T given by f (1) = 3, f (2) = 1, f (3) = 2, as (a) a 2 × 3 array, (b) a crossing diagram. Find its swapping number and sign. Definition 39 Let f : T → T and g : T → T be two permutations. We can compose them to get another permutation, the composition, denoted f g : T → T: t 7−→ f (t) 7−→ g(f (t)) T → T → T Notation We shall follow standard convention and write our compositions of permutations left-to-right. (This is of course in contrast to the right-to-left composition of functions you may have seen in calculus.) When a possible ambiguity may arise, we call this type of composition ”composition as permutations” and call ”right-to-left composition” the ”composition as functions”. When f = g then we write f f as f 2 . In general, we write the n-fold composition f...f (n times) as f n . Every permutation f has the property that there is some integer N > 0, which depends on f , such that f N = 1. (In other words, if you repeatedly apply a permutation enough times you will eventually obtain the identity permutation.) Definition 40 The smallest integer N > 0 such that f N = 1 is called the order of f . Example 41 Let T = {1, 2, 3} and let Ã

f= We have

Ã

fg =

1 2 3 2 1 3 1 2 3 1 3 2

!

Ã

g=

1 2 3 3 1 2

!

!

f 2 = 1,

,

g 3 = 1.

Exercise 3.0.3 Compute (a) f g and (b) the order of f and the order of g, where à ! à ! 1 2 3 1 2 3 (a) f= g= 3 2 1 3 1 2 Ã

(b)

f=

1 2 3 3 1 2

!

Ã

g=

1 2 3 2 1 3

!

36

CHAPTER 3. PERMUTATIONS

Exercise 3.0.4 If f, g, h are permutations of T , is (f g)h = f (gh)? Explain why.

3.1

Inverses

We can look at the graph of a function f : T → T and determine (a) if it is injective, (b) if it is surjective, (c) the inverse f −1 , if it exists. Indeed, from the graph of f we can determine the image f (T ) and this determines if f is surjective or not. The inverse exists only if f is injective (hence, in our case, surjective by exercise 2.1.3). It’s graph is determined by reflecting the graph of f about the diagonal, x=y.

Lemma 42 The following statements are equivalent: (1) f : T → T is injective, (2) f : T → T is surjective, (3) |f (T )| = |T |.

proof: The equivalence of the first two statements is by the exercise at the end of chapter 1. (2) is equivalent to (3), by definition of surjectivity. 2

Example 43 The inverse of Ã

is obtained by reflecting the graph

1 2 3 3 1 2

!

3.1. INVERSES

about the x = y line:

37

38

CHAPTER 3. PERMUTATIONS

Exercise 3.1.1 Graph and determine the inverses of the following permutations: Ã ! 1 2 3 (a) f= 2 1 3 Ã

(b)

f= Ã

(c)

f=

1 2 3 4 2 3 4 1

!

1 2 3 4 5 2 1 5 3 4

!

There are two more commonly used ways of expressing a permutation. The first is the ”matrix notation”: Definition 44 To a permutation f : T → T , given by Ã

f=

1 2 ... n f (1) f (2) ... f (n)

!

3.1. INVERSES

39

we associate to it the matrix P (f ) of 00 s and 10 s defined as follows: the ij-th entry of P (f ) is 1 if j = f (i) and is 0 otherwise. (A brief introduction to matrices is given in the appendix.) Definition 45 A square matrix which has exactly one 1 per row and per column (as P (f ) does) is called a permutation matrix . Lemma 46 There are n! distinct n × n permutation matrices and there are n! distinct permutations of the set {1, 2, ..., n}. Exercise 3.1.2 Prove this using the multiplication counting principle from the section on counting in the previous chapter. Example 47 The matrix of the permutation f given by Ã

1 2 3 2 1 3

f= is

!





0 1 0  P (f ) =  1 0 0   0 0 1

Note that matrix multiplication gives 









0 1 0 1 2       1 0 0  2  =  1  0 0 1 3 3 from which we can recover the 2 × 3 array. Theorem 48 If f : T → T is a permutation then 

(a)

  P (f )   

1 2 .. . n





    =    

f (1) f (2) .. .

     

f (n)

Furthermore, the inverse of the matrix of the permutation is the matrix of the inverse of the permutation: (b) P (f )−1 = P (f −1 ), and the matrix of the product is the product of the matrices: (c) P (f g) = P (f )P (g).

40

CHAPTER 3. PERMUTATIONS

proof: If ~v is the column vector with entries v1 , v2 , ..., vn (the vi are arbitrary real numbers) then P (f )~v is the column vector whose ith coordinate is equal to vj if f sends i to j, by definition of P (f ). Since, in this case, j = f (i) (here we write f (i) to denote the image of i under the permutation f , ’ even though i really gets plugged into f on the left), this implies that P (f )~v is the column vector with entries vf (1) , vf (2) , ..., vf (n) . This proves (a). Note (b) is a consequence of (c) so we need only prove (c). We compute P (f g)~v and P (f )P (g)~v . By the same reasoning as in (a), we find that the ith coordinate of P (f g)~v is v(f g)(i) . Similarly, the ith coordinate of P (g)~v is vi0 = vg(i) . Therefore, the ith coordinate of P (f )(P (g)~v ) is vf0 (i) = vg(f (i)) = v(f g)(i) . This implies P (f g)~v = P (f )P (g)~v . Since the vi were arbitrary real numbers, this implies the theorem. 2 Example 49 Let Ã

f=

1 2 3 2 1 3

!

Ã

,

g=

1 2 3 3 2 1

!

Ã

,

h=

1 2 3 2 3 1

!

,

so f = f −1 , g = g −1 , h = f g. Moreover, 



0 0 1   P (g) =  0 1 0  , 1 0 0





0 1 0   P (h) =  0 0 1  . 1 0 0

A direct matrix calculation verifies that P (f )P (g) = P (f g) = P (h) and P (h−1 ) = P (g −1 f −1 ) = P (g −1 )P (f −1 ) = P (g)P (f ), as predicted by the above theorem. The matrix can be determined from the graph of the function f : T → T as follows: in the n × n grid of integral points (x, y), with x and y integers between 1 and n inclusive, fill in all the plotted points with 10 s and all the unplotted points with 00 s The resulting n × n array is the matrix P (f ). Rubik’s cubers will often, without knowing it perhaps, use the following lemma to solve their cube: Lemma 50 Let r ∈ Sn denote any permutation and let i, j denote distinct integers belonging to {1, 2, ..., n}. Let s denote the permutation sending i to j: (i)s = j.

3.2. CYCLE NOTATION

41

Then sr = r−1 sr is the permutation sending (i)r to (j)r: sr ((i)r) = (j)r. More specifically (and this is the specific case which this lemma is most oftened applied): let i1 , i2 , ..., ik denote distinct integers belonging to {1, 2, ..., n}. Let s denote the permutation sending ij to ij+1 : s(ij ) = ij+1 ,

1 ≤ j < k,

s(ik ) = i1 ,

s(m) = m, ∀m ∈ / {i1 , ..., ik }.

Then sr = r−1 sr is the permutation sending (ij )r to (ij+1 )r: sr ((ij )r) = (ij+1 )r,

1 ≤ j < k,

sr ((ik )r) = (i1 )r,

sr (m) = m, ∀m ∈ / {(i1 )r, ..., (ik )r}.

In other words, if you have a move s which is a 3-cycle on 3 particular edges, say uf 7−→ ul 7−→ ur 7−→ uf, and another move r which sends these edges somewhere else, say r = F 2 so that r : uf 7−→ df but leaves the other edges alone, then r−1 sr is the permutation df 7−→ ul 7−→ ur 7−→ df. Try it! A proof of this lemma will be given in chapter 8.

3.2

Cycle notation

The most common notation for a permutation is probably the ”cycle notation”. The symbol (a1 a2 ... ar )

(some r less than or equal to n)

denotes the permutation f of T which is defined by f (a1 ) = a2 , f (a2 ) = a3 , ..., f (ar ) = a1 , and f (i) = i, if i is not equal to one of the a1 , ..., ar . Such a permutation is called cyclic. The number r is called the length of the cycle. We call two such cycles (a1 a2 ... ar ) and (b1 b2 ... bt ) disjoint if the sets {a1 , a2 , ..., ar } and {b1 , b2 , ..., bt } are disjoint.

42

CHAPTER 3. PERMUTATIONS

Lemma 51 If f and g are disjoint cyclic permutations of T then f g = gf . proof: This is clear since the permutations f and g of T affect disjoint collections of integers, so the permutations may be performed in either order. 2 Lemma 52 The cyclic permutation (a1 a2 ... ar ) has order r. proof: Note f (a1 ) = a2 , f 2 (a1 ) = a3 , ..., f r−1 (a1 ) = ar , f r (a1 ) = a1 , by definition of f . Likewise, for any i = 1, ..., r, we have f r (ai ) = ai . 2 Theorem 53 Every permutation f : T → T is the product of disjoint cyclic permutations. More precisely, if f is a permutation of {1, 2, ..., n} (with n > 1) then there are non-empty disjoint subsets of distinct integers S1 = {a11 , ..., a1,r1 } ⊂ {1, 2, ..., n}, S2 = {a21 , ..., a2,r2 } ⊂ {1, 2, ..., n}, ..., Sk = {ak1 , ..., ak,rk }, such that {1, 2, ..., n} = S1 ∪ ... ∪ Sk ,

n = r1 + r2 + ... + rk ,

and f = (a11 , ..., a1,r1 )...(ak1 , ..., ak,rk ). This product is called a cycle decomposition of f . If we rearrange the cardinalities ri of these sets Si in decreasing order, say we write this as r10 ≥ r20 ≥ ... ≥ rk0 , then the k-tuple (r10 , ..., rk0 ) is called the cycle structure of f and f is called a (r10 , ..., rk0 )-cycle. For example, (1, 2)(3, 4, 5) is a (3, 2)-cycle. proof: The proof is constructive. Let f : T → T be a permutation. List all the elements {c10 = 1, c11 = f (1), c12 = f 2 (1), c13 = f 3 (1), ...},

3.2. CYCLE NOTATION

43

(which must, of course, be finite in number but might also only contain the single element c10 = 1). This is called the ”orbit of 1 under f ”. Now list the elements in the ”orbit of 2”: {c20 = 2, c21 = f (2), c22 = f 2 (2), c23 = f 3 (2), ...}, and so on until we get to the ”orbit of n”: {cn0 = 2, cn1 = f (n), cn2 = f 2 (n), cn3 = f 3 (n), ...}. If you pick any two of these n sets, they will either be the same (up to ordering) or disjoint. Denote all the distinct orbits which contain at least two elements by O1 , ..., Ok . (It doesn’t matter what order you write them in or in what order you write the elements in each individual orbit.) Suppose that O1 = a11 , ..., a1,r1 so |O1 | = r1 , O2 = a21 , ..., a2,r2 so |O2 | = r2 , . . . Ok = ak1 , ..., ak,rk so |Ok | = rk . In this case, r1 + r2 + ... + rk ≤ n, with equality if and only if none of the orbits is a singleton. The cycle notation of f is the expression (a11 a12 ...a1,r1 )...(ak1 ak2 ... ak,rk ). 2 Example 54

• The cycle notation for Ã

1 2 3 2 1 3

!

is (1 2). • The cycle notation for

is (1 2 3).

Ã

1 2 3 2 3 1

!

44

CHAPTER 3. PERMUTATIONS • The cycle notation for

Ã

1 2 3 4 3 4 1 2

!

is (1 3)(2 4) = (2 4)(1 3). • The cycle notation for Ã

1 2 3 4 5 3 4 1 5 2

!

is (1 3)(2 4 5) = (4 5 2)(1 3). • The disjoint cycle decomposition of (2, 3, 7)(3, 7, 10) is (2, 3)(7, 10). Exercise 3.2.1 Divide a square into 4 subsquares (”facets”) and label them 1, 2, 3, 4. For example, ----------------| | | | 1 | 2 | | | | ----------------| | | | 3 | 4 | | | | ----------------Let r denote counterclockwise rotation by 90 degrees. Then, as a permutation on the facets, r = (1 3 4 2). Let fx denote reflection about the horizonal line dividing the square in two, let fy denote reflection about the vertical line dividing the square in two. Use the cycle notation to determine the permutations of the facets (a) r2 (b) r3 , (c) fx , (d) fy , (e) fx ∗ r ∗ fx , (f ) fx ∗ fy . Exercise 3.2.2 Label the 24 facets of the 2 × 2 Rubik’s cube as follows:

3.2. CYCLE NOTATION

45

+--------------+ | 1 2 | |

u

|

| 3 4 | +--------------+--------------+--------------+--------------+ | 5 6 | 9 10 | 13 14 | 17 18 | |

l

|

f

|

r

|

b

|

| 7 8 | 11 12 | 15 16 | 19 20 | +--------------+--------------+--------------+--------------+ | 21 22 | |

d

|

| 23 24 | +--------------+ (You may want to xerox this page then cut this cube out and tape it together for this exercise.) Let X denote rotation clockwise by 90 degrees of the face labeled x, where x ∈ {r, l, f, b, u, d} (so, for example, if x = f then X = F ). Use the cycle notation to determine the permutations of the facets given by (a) R, (b) L, (c) F , (d) B, (e) U , (f ) D. Lemma 55 A cyclic permutation is even if and only if the length of its cycle is odd. A general permutation f : T → T is odd if and only if the number of cycles of even length in its cycle decomposition is odd.

46

CHAPTER 3. PERMUTATIONS

We shall not prove this here. (For a proof, see Theorem 3.3 in Gaglione [G], §3.2.) Lemma 56 The order of a permutation is the least common multiple (lcm) of the lengths r1 , r2 , ..., rk of the disjoint cycles in its cycle decomposition. Example 57 The order of (1 3)(2 4) is 2. It is even. The order of (1 3)(2 4 5) is 6. It is odd.

3.3

An algorithm to list all the permutations

In Martin Gardner’s Scientific American article [Gar1] an algorithm is mentioned which lists all the permutations of {1, 2, ..., n}. This algorithm, due originally to the mathematician Hugo Steinhaus, gives the fastest known method of listing all permutations of {1, 2, ..., n}. We shall denote each permutation by the second row in its 2 × n array notation. For example, in the case n = 2: 1 2 2 1 are the permutations. To see the case n = 3, the idea is to (a) write down each row n = 3 times each as follows: 1 1 1 2 2 2

2 2 2 1 1 1

(b) ”weave” in a 3 as follows 1 1 3 3 2 2

2 3 1 2 3 1

3 2 2 1 1 3

3.3. AN ALGORITHM TO LIST ALL THE PERMUTATIONS In case n = 4, the idea is to

(a) write down each row n = 4 times each as follows:

1 1 1 1 1 1 1 1 3 3 3 3 3 3 3 3 2 2 2 2 2 2 2 2

2 2 2 2 3 3 3 3 1 1 1 1 2 2 2 2 3 3 3 3 1 1 1 1

3 3 3 3 2 2 2 2 2 2 2 2 1 1 1 1 1 1 1 1 3 3 3 3

47

48

CHAPTER 3. PERMUTATIONS (b) now ”weave” a 4 in: 1 1 1 4 4 1 1 1 3 3 3 4 4 3 3 3 2 2 2 4 4 2 2 2

2 2 4 1 1 4 3 3 1 1 4 3 3 4 2 2 3 3 4 2 2 4 1 1

3 4 2 2 3 3 4 2 2 4 1 1 2 2 4 1 1 4 3 3 1 1 4 3

4 3 3 3 2 2 2 4 4 2 2 2 1 1 1 4 4 1 1 1 3 3 3 4

In general, we have the following Theorem 58 (Steinhaus) There is a sequence of 2-cycles, (a1 , b1 ),...,(aN , bN ), where N = n! − 1, such that each non-trivial permutation f of {1, 2, ..., n} is of the form f=

k Y

(ai , bi ),

i=1

for some k, 1 ≤ k ≤ N . In other words, each permutation may be written as a product of (not necessarily disjoint) 2-cycles. This will be proven in the section on Campanology below.

3.3. AN ALGORITHM TO LIST ALL THE PERMUTATIONS

49

There is an analogous result valid only for even permutations: each even permutation may be written as a product of (not necessarily disjoint) 3cycles. This will be stated more precisely (and proved) later - see Proposition 158.

50

CHAPTER 3. PERMUTATIONS

Chapter 4 Permutation Puzzles ”How can it be that mathematics, being after all a product of human thought independent of experience, is so admirably adapted to the objects of reality?” Albert Einstein ”Though this be madness, yet there is method in’t.” Shakespeare We shall describe several permutation puzzles in this chapter. A one person game is a sequence of moves following certain rules satisfying • there are finitely many moves at each stage, • there is a finite sequence of moves which yields a solution, • there are no chance or random moves, • there is complete information about each move, • each move depends only on the present position, not on the existence or non-existence of a certain previous move (such as chess, where castling is made illegal if the king has been moved previously). A permutation puzzle is a one person game (solitaire) with the following five properties listed below. Before listing the properties, we define the ”puzzle position” to be the set of all possible legal moves. The five properties of a permutation puzzle are: 51

52

CHAPTER 4. PERMUTATION PUZZLES 1. for some n > 1 depending only on the puzzle’s construction, each move of the puzzle corresponds to a unique permutation of the numbers in T = {1, 2, ..., n},

2. if the permutation of T in (1) corresponds to more than one puzzle move then the the two positions reached by those two respective moves must be indistinguishable,

3. each move, say M , must be ”invertible” in the sense that there must exist another move, say M −1 , which restores the puzzle to the position it was at before M was performed,

4. if M1 is a move corresponding to a permutation f1 of T and if M2 is a move corresponding to a permutation f2 of T then M1 ∗ M2 (the move M1 followed by the move M2 ) is either

• not a legal move, or • corresponds to the permutation f1 ∗ f2 .

Notation: As in step 4 above, we shall always write successive puzzle moves left-to-right.

4.1

15 puzzle

One of the earliest and most popular permutation puzzles is the ”15 puzzle”. The ”solved position” looks like

4.1. 15 PUZZLE

53

These numbered squared represent sliding blocks which can only move into the blank square. We shall sometimes label the blank square as ”16” for convenience. The moves of the puzzle consist of sliding numbered squares (such as 12, for example) into the blank square (e.g., swapping 12 with 16). In this way, each move of this puzzle may be regarded as a permutation of the integers in 1, 2,..., 16.

Exercise 4.1.1 Check that the five conditions of a permutation puzzle are satisfied by the 15 puzzle.

Not every permutation of the {1, 2, ..., 16} corresponds to a possible position of the puzzle. For example, the position

54

CHAPTER 4. PERMUTATION PUZZLES

cannot be attained from the previous position. (The mathematical reason for this is explained in 6.3 below, for example.) Apparently, the puzzle inventor Sam Loyd applied for a U.S. patent for the above puzzle (the one with the 14, 15 swapped) but since it could not be ”solved” - i.e., put in the correct order 1, 2, ..., 15 - no working model could be supplied, so his patent was denied. (This is in spite of the fact that there were apparently thousands of them on the market already.)

The moves of the 15 puzzle may be denoted as follows: suppose we are in a position such as

4.2. DEVIL’S CIRCLES (OR HUNGARIAN RINGS)

55

The possible moves are R = Ru,r,d,l = (r 16) = swap r and 16, L = Lu,r,d,l = (l 16) = swap l and 16, U = Uu,r,d,l = (u 16) = swap u and 16, D = Du,r,d,l = (d 16) = swap d and 16. Exercise 4.1.2 Verify that the five defining properties of a permutation puzzle are satisfied by this example. We shall call the 15 puzzle a planar puzzle since all its pieces lie on a flat board.

4.2

Devil’s circles (or Hungarian rings)

This is a planar puzzle consisting of two or more interwoven ovals, each of which has several labeled (by colors or numbers) pieces, some of which may

56

CHAPTER 4. PERMUTATION PUZZLES

belong to more than one oval. A puzzle move consists of shifting an oval by one or more ”increments”, and hence all the pieces on it, along the oval’s grooved track. The pieces are equally spaced apart (in spite of the typed depiction below) and those pieces which lie on more than one oval can be moved along either oval. For simplicity, consider the puzzle consisting of only two ovals, each having 6 pieces:

The pieces 1 and 3 can be moved along either oval. Note that each move corresponds to a unique permutation of the numbers in {1, 2, ..., 10}. For example, rotating the right-hand oval clockwise one increment corresponds to the permutation 1 2 3 4 5 6 7 8 9 10 6 1 2 3 4 5 7 8 9 10 which we may write in cycle notation as (6 5 4 3 2 1). Exercise 4.2.1 Verify that the five defining properties of a permutation puz-

4.3. EQUATOR PUZZLE

57

zle are satisfied by this example.

4.3

Equator puzzle

This puzzle is in the shape of a sphere but has 3 circular bands encircling a sphere, each having 12 square-shapped pieces and each band intersecting each other at a 90 degree angle. Each pair of circles intersects at two points, or ”nodes”, and at each such node there is a puzzle piece shared by the two circular bands. There are 6 nodes total. The total number of movable pieces is therefore 3 · 12 − 6 = 30. On some puzzles the sphere is painted a map of the earth, others have colored puzzle pieces. The rough idea, minus any colors, is depicted in the following picture:

For concreteness, suppose we are looking at a globe equator puzzle. The longitudinal band circles the equator, one latitudinal band passes through North America and the other latitudinal band passes through Europe and

58

CHAPTER 4. PERMUTATION PUZZLES

Africa. A move of the puzzle consists of rotating one of the bands (along with all the pieces it contains) in either direction. Successive puzzle moves may change the ”orientation” of a piece, as we will see later. When the 30 pieces are such that the puzzle is a correct map of the Earth then we call the position ”solved”. For ease of drawing, let us redraw the globe of the Earth using the ”Mercatur projection”, i.e., as a wall map: 1 2 3 4 5 6 7

23

24

1 13 14 15 16 17 7

25

26

1 12 11 10 9 8 7

27

28

1 22 21 20 19 18 7

29

30

The "solved" position Sometimes we shall also denote 1 by NP (”north pole”) and 7 by SP (”south pole”). This description is a little misleading due to the fact that it tells us where a piece is but not, for example, whether it is upside down or not. We shall ignore this problem for now and simply describe how a move affects the position of a piece. We see from the above labeling that any move of the Equator puzzle corresponds to a unique permutation of the integers in 1, 2, ..., 30. For example, the move which rotates the equator east-to-west by 30 degrees corresponds to the permutation (4 30 29 20 28 27 10 26 25 15 24 23). Now we shall show to assign an orientation to a piece. We shall regard an orientation (which is, after all, simply an indication of what angle the piece is ”twisted”) as an integer 0, 1 ,2, or 3. First, if a piece is not in its correct position, it gets an orientation of 0. If a piece is in its correct position then it gets a 0, 1, 2, or 3, depending on its angle from the correct angle (i.e., the angle the piece has in the ”solved” position):

4.3. EQUATOR PUZZLE

59

0 | | | 3 ------------------- 1 | | | 2 Example 59 A piece which has been rotated by 90 degrees counterclockwise from its correct orientation gets an orientation of 3. In general, the labels for the pieces of the Equator puzzle should be choosen from the set given by the Cartesian product of the set of integers used to label the positions, {1, 2, ..., 30}, by the set of integers used to label the orientations: S = {1, 2, ..., 30} × {0, 1, 2, 3} = {(m, n) | 1 ≤ m ≤ 30, 0 ≤ n ≤ 3}. Each move of the Equator corresponds to a unique permutation of the set S. There are 120 elements of S, call them S = {s1 , s2 , ..., s120 }. If we identify the set S with the set T = {1, ..., 120} then we move of the Equator corresponds to a unique permutation of the set T . Exercise 4.3.1 Verify that the Equator puzzle satisfies the five defining properties of a permutation puzzle. Question: Can you show the following: If a piece is correctly oriented then its antipodal piece is also correctly oriented? Notation: We introduce notation for 3 basic moves of the Equator puzzle which generate all possible puxzzle moves. Let us label the three circular

60

CHAPTER 4. PERMUTATION PUZZLES

bands on the globe as C1, C2, and C3. Let C1 be the band which, in the solved position, contains the pieces labeled 1, 2, ..., 12; let C2 be the band which, in the solved position, contains the pieces labeled 7, 13, ..., 22; and let C3 be the band which, in the solved position, contains the pieces on the equator. Let r1 be the puzzle move associated to the rotation of C1 given by 1 2 3 4 5 6 7 8 9 10 11 12 2 3 4 5 6 7 8 9 10 11 12 1 Let r2 be the puzzle move associated to the rotation of C2 given by 1 13 14 15 16 17 17 18 19 20 21 22 13 14 15 16 17 17 18 19 20 21 22 1 Let r3 be the puzzle move associated to the rotation of C3 given by 4 23 24 15 25 26 10 27 28 20 29 30 30 4 23 24 15 25 26 10 27 28 20 29 It is clear after a little thought that each of these moves corresponds to a permutation of the 120 position/orientation labels of the pieces of the equator puzzle. Furthermore, such a permutation determines the puzzle position uniquely since it specifies the facet’s position and orientation. Exercise 4.3.2 Verify that the remaining properties of a permutation puzzle are satisfied.

4.4

Rainbow Masterball

Some rules for the rainbow masterball (referred to simply as ”masterball” in the following): A masterball sphere has 32 tiles of 8 distinct colors. We shall assume that the masterball is in a fixed position in space, centered at the origin. A geodesic path from the north pole to the south pole is called a longitudinal line and a closed geodesic path parallel to the equator is called a latitudinal line. There are 8 longitudinal lines and 3 latitudinal lines. In spherical coordinates, the longitudinal lines are at the angles which are multiples of π/4 (i.e., at θ = nπ/4, n = 1, .., 8) and the latitudinal lines are at φ = π/4, π/2, 3π/4. (Here π = 3.141592... as usual.)

4.4. RAINBOW MASTERBALL

61

The sphere shall be oriented by the right-hand rule - the thumb of the right hand wrapping along the polar axis points towards the north pole. We assume that one of the longitudinal lines has been fixed once and for all. This fixed line shall be labeled ”1”, the next line (with respect to the orientation above) as ”2”, and so on. Allowed moves: One may rotate the masterball east-to-west by multiples of π/4 along each of the 4 latitudinal bands or by multiples of π along each of the 8 longitudinal lines. A facet will be one of the 32 subdivisions of the masterball created by these geodesics. A facet shall be regarded as immobile positions on the sphere and labeled either by an integer i ∈ {1, ..., 32} or by a pair (i, j) ∈ [1, 4]×[1, 8], whichever is more convenient at the time. If a facet has either the north pole or the south pole as a vertex then we call it a small (or polar) facet. Otherwise, we call a facet large (or middle or equatorial). A coloring of the masterball will be a labeling of each facet by one of the 8 colors in such a way that (a) each of the 8 colors occurs exactly twice in the set of the 16 small

62

CHAPTER 4. PERMUTATION PUZZLES

facets, (b) each of the 8 colors occurs exactly twice in the set of the 16 large facets. A move of the masterball will be a change in the coloring of the masterball associated to a sequence of manuevers as described above. If we now identify each of the 8 colors with an integer in {1, ..., 8} and identify the collection of facets of the masterball with a 4×8 array of integers in this range. To solve an array one must, by an appropriate sequence of moves corresponding to the above described rotations of the masterball, put this array into a ”rainbow” position so that the matrix entries of each column has the same number. Thus the array 1 1 1 1

2 2 2 2

3 3 3 3

4 4 4 4

5 5 5 5

6 6 6 6

7 7 7 7

8 8 8 8

6 1 1 1

7 2 2 2

8 3 3 3

1 4 4 4

2 5 5 5

3 6 6 6

4 7 7 7

5 8 8 8

is ”solved”. The array

corresponds to a rotation of the north pole facets by 3π/4. Notation We use matrix notation to denote the 32 facets of the masterball. The generators for the latitudinal rotations are denoted r1 , r2 , r3 , r4 . For example, r1 sends

to

which is pictured as:

11 21 31 41

12 22 32 42

13 23 33 43

14 24 34 44

15 25 35 45

16 26 36 46

17 27 37 47

18 28 38 48

12 21 31 41

13 22 32 42

14 23 33 43

15 24 34 44

16 25 35 45

17 26 36 46

18 27 37 47

11 28 38 48

4.4. RAINBOW MASTERBALL

63

As you look down at the ball from the north pole, this move rotates the ball clockwise. The other moves r2 , r3 , r4 rotate the associated band of the ball in the same direction - clockwise as viewed from the north pole. The generators for the longitudinal rotations are denoted f1 , f2 , ..., f8 . For example, f1 sends 11 21 31 41

12 22 32 42

13 23 33 43

14 24 34 44

15 25 35 45

16 26 36 46

17 27 37 47

18 28 38 48

44 34 24 15

43 33 23 14

42 32 22 13

41 31 21 12

16 25 35 45

17 26 36 46

18 27 37 47

11 28 38 48

to

which is pictured as:

64

CHAPTER 4. PERMUTATION PUZZLES

With these rules, one can check the relation f5 = r14 ∗ r24 ∗ r34 ∗ r44 ∗ f1 ∗ r14 ∗ r24 ∗ r34 ∗ r44 . Exercise 4.4.1 Find similar identities for f6 , f7 , f8 . Also, one can check that r1 = (f3 ∗ f7 )−1 ∗ r4−1 ∗ f3 ∗ f7 . Exercise 4.4.2 There are similar identities for r2 , r3 , r4 . Find them. Identify the facets of the masterball with the entries of the array 8 7 6 5 4 3 2 1 16 15 14 13 12 11 10 9 24 23 22 21 20 19 18 17 32 31 30 29 28 27 26 25

4.5. 2 × 2 RUBIK’S CUBE

65

(there is a reason for labeling the facets ”backwards” like this but it’s not important). We may express the generators of the masterball group in disjoint cycle notation as a subgroup of S32 (the symmetric group on 32 letters):

r1−1 = (1 2 3 4 5 6 7 8), −1 r2 = (9 10 11 12 13 14 15 16), r3−1 = (17 18 19 20 21 22 23 24), r4−1 = (25 26 27 28 29 30 31 32), f1 = (5 32)(6 31)(7 30)(8 29)(13 24)(14 23)(15 22)(16 21), f2 = (4 31)(5 30)(6 29)(7 28)(12 23)(13 22)(14 21)(15 20), f3 = (3 30)(4 29)(5 28)(6 27)(11 22)(12 21)(13 20)(14 19), f4 = (2 29)(3 28)(4 27)(5 26)(10 21)(11 22)(12 23)(13 24), f5 = (1 28)(2 27)(3 26)(4 25)(9 20)(10 19)(11 18)(12 17), f6 = (8 27)(1 26)(2 25)(3 32)(16 19)(9 18)(10 17)(11 24), f7 = (7 26)(8 25)(1 32)(2 31)(15 18)(16 17)(9 24)(10 23), f8 = (6 25)(7 32)(8 31)(1 30)(14 17)(15 24)(16 23)(9 22),

Exercise 4.4.3 Verify that the properties of a permutation puzzle are satisfied for this puzzle.

More information on this puzzle will be given in a later chapter.

4.5

2 × 2 Rubik’s cube

The ”pocket” Rubik’s cube has 6 sides, or ”faces”, each of which has 2 · 2 = 4 ”facets”, for a total of 24 facets:

66

CHAPTER 4. PERMUTATION PUZZLES

Fix an orientation of the Rubik’s cube in space. Therefore, we may label the 6 sides as f, b, l, r, u, d, as in the picture. It has 8 subcubes. Each face of the cube is associated to a ”slice” of 4 subcubes which share a facet with the face. The face, along with all of the 4 cubes in the ”slice”, can be rotated by 90 degrees clockwise. We denote this move by the upper case letter associated to the lower case letter denoting the face. For example, F denotes the move which rotates the front face by 90 degrees to clockwise.

4.5. 2 × 2 RUBIK’S CUBE

67

As in chapter 4, we label the 24 facets of the 2×2 Rubik’s cube as follows:

68

CHAPTER 4. PERMUTATION PUZZLES +--------------+ | 1 2 | |

u

|

| 3 4 | +--------------+--------------+--------------+--------------+ | 5 6 | 9 10 | 13 14 | 17 18 | |

l

|

f

|

r

|

b

|

| 7 8 | 11 12 | 15 16 | 19 20 | +--------------+--------------+--------------+--------------+ | 21 22 | |

d

|

| 23 24 | +--------------+ The 24 facets will be denoted by xyz where x is the face on which the facet lives and y, z (or z, y - it doesn’t matter) indicate the 2 edges of the facet. Written in clockwise order: front back right left up down

face: face: face: face: face: face:

fru, blu, rbu, lfu, urb, drf,

frd, bld, rbd, lfd, urf, drb,

fld, brd, rfd, lbd, ulf, dlb,

flu bru rfu lbu ulb dlf

Exercise 4.5.1 Verify that the properties of a permutation puzzle are satisfied for this puzzle. For future reference, we call this system of notation (which we will also use for the 3 × 3 and 4 × 4 Rubik’s cube) the Singmaster notation.

4.6. 3 × 3 RUBIK’S CUBE

4.6

69

3 × 3 Rubik’s cube

In this section we shall, for the most part, simply introduce enough notation (due to Singmaster [Si]) to allow us to check that the puzzle is in fact a permutation puzzle. We shall also introduce a two-person game which is easier to play and learn than solving the cube. The Rubik’s cube has 6 sides, or ”faces”, each of which has 3 · 3 = 9 ”facets”, for a total of 54 facets. We label these facets 1, 2, ..., 54 as follows: +--------------+ | 1 2 3 | |

4

u

5 |

| 6 7 8 | +--------------+--------------+--------------+--------------+ | 9 10 11 | 17 18 19 | 25 26 27 | 33 34 35 | | 12

l

13 | 20

f

21 | 28

r

29 | 36

b

37 |

| 14 15 16 | 22 23 24 | 30 31 32 | 38 39 40 | +--------------+--------------+--------------+--------------+ | 41 42 43 | | 44

d

45 |

| 46 47 48 | +--------------+ then the generators, corresponding to the six faces of the cube, may be written in disjoint cycle notation as: F = (17 19 24 22)(18 21 23 20)(6 25 43 16)(7 28 42 13)(8 30 41 11), B = (33 35 40 38)(34 37 39 36)(3 9 46 32)(2 12 47 29)(1 14 48 27), L = (9 11 16 14)(10 13 15 12)(1 17 41 40)(4 20 44 37)(6 22 46 35), R = (25 27 32 30)(26 29 31 28)(3 38 43 19)(5 36 45 21)(8 33 48 24), U = (1 3 8 6)(2 5 7 4)(9 33 25 17)(10 34 26 18)(11 35 27 19), D = (41 43 48 46)(42 45 47 44)(14 22 30 38)(15 23 31 39)(16 24 32 40).

70

CHAPTER 4. PERMUTATION PUZZLES

Exercise 4.6.1 Check this. (It is helpful to xerox the above diagram, cut it out and tape together a paper cube for this exercise.) The notation for the facets will be similar to the notation used for the 2 × 2 Rubik’s cube. The corner factes will have the same notation and the edge facets will bve denoted by xy, where x is the face the facet lives on and y is the face the facet borders to. In clockwise order, starting with the upper right-hand corner of each face: front back right left up down

face: face: face: face: face: face:

fru, blu, rbu, lfu, urb, drf,

fr, bl, rb, lf, ur, dr,

frd, bld, rbd, lfd, urf, drb,

fd, bd, rd, ld, uf, db,

fld, brd, rfd, lbd, ulf, dlb,

fl, br, rf, lb, ul, dl,

flu, bru, rfu, lbu. ulb, dlf,

fu bu ru lu ub df

Exercise 4.6.2 Verify that the properties of a permutation puzzle are satisfied for this puzzle.

4.6.1

The superflip game

The position of the Rubik’s cube where every edge is flipped, but all the others subcubes are unaffected, is called the superflip position. To play the game, first choose two particular faces as your up (U) and front (F) face - say white is up and red is front (assuming you have a cube with adjacent white and red faces). Imagine the cube being placed in space with rectangular coordinate axes in such a way that the bdl corner is at the origin (0, 0, 0), the dl edge is along the x-axis and the bl edge is along the z-axis. The rules (”slice-superflip game”): 1. Players alternate making moves starting with the cube in the solved position. The first player is determined by (say) a coin toss. 2. A move consists of flipping over exactly two edges. Both edges must lie in a slice. The edge closest to the origin (or, if this is a tie, closest to the x-axis) must be flipped from ”solved” to ”wrong”. 3. The first player to reach the superflip position wins.

4.7. 4 × 4 RUBIK’S CUBE

71

This game is related to a game which might be called a ”three dimensional acrostic twins game”. (See [BCG], vol II, page 441, for a two dimensional acrostic twins game.) Several alternate versions of this game may also be played. ”Nonslice-superflip game”: The rules are the same except the condition that the two edges belong to the same slice is either dropped altogether or replaced by the condition that the two edges do not belong to the same slice. ”M¨obius-superflip game”: The rules for this version are the same except the condition that two edges are flipped is to be replaced by any number of edges less than 6 (i.e., exactly 2, 4, or 6) is to be flipped. This game is related to a game which might be called a ”three dimensional M¨obius”. (See [BCG], vol II, page 434, for a M¨obius game.)

4.7

4 × 4 Rubik’s cube

The 4 × 4 Rubik’s cube has 6 sides, or ”faces”, each of which has 4 · 4 = 16 ”facets”, for a total of 96 facets. As usual, we fix an orientation of the cube in space, so we may pick a front face, back face, ... . We label these facets 1, 2, ..., 96 as follows:

72

CHAPTER 4. PERMUTATION PUZZLES +-----------------+ | 49

50

51

52

|

| 61

62

64

|

| 73

63 u 74 75

76

|

| 77

78

80

|

79

+------------------+-----------------+-----------------+-----------------+ |

53

54

55

56

|

1

2

3

4

|

65

66

68

| 13

14

77

67 l 78 79

80

| 25

26

15 f 27

| |

89

90

92

| 37

38

39

91

|

5

6

7

8

|

9

10

11

12 |

16

| 17

18

20

|

21

22

| 29

30

32

|

33

34

23 b 35

24 |

28

19 r 31

40

| 41

42

43

44

|

45

46

47

48 |

36 |

+------------------+-----------------+-----------------+-----------------+ | 57

58

59

60

|

| 69

70

72

|

| 81

71 d 82 83

84

|

| 93

94

96

|

95

+-----------------+

The reader may want to xerox the above diagram, cut it out and tape together a paper cube. Notation: We need notation for the facets and for the moves. Facets: To label the facets, we must pick an orientation of each face, say clockwise. For example, the the front face may be labeled as

4.7. 4 × 4 RUBIK’S CUBE

73

+---------------------+ | flu

fu1

fu2

fru

|

| fl1

f4

f1

fr1

|

| fl2

f3

f2

fr2

|

| fld

fd2

fd2

frd

|

f

+---------------------+

The labeling of the other faces is similar.

Exercise 4.7.1 Label the other 5 faces.

Moves: Parallel to each face x are 4 slices of 16 subcubes each labeled X1 , X2 , X3 , X4 , in order of their distance from the face. For example, the front face f has 16 subcubes comprising the F1 slice, the two inner slices are F2 , F3 , and the last slice F4 is actually the same as the first slice B1 associated to the back face.

74

CHAPTER 4. PERMUTATION PUZZLES

The 12 generators (written in disjoint cycle notation), corresponding 2 each to the six faces of the cube are given by: U1 = (49 52 88 85)(62 63 75 74)(50 64 87 73)× ×(51 76 86 61)(5 1 53 9)(6 2 54 10)(7 3 55 11)(8 4 56 12), U2 = (17 13 65 21)(18 14 66 22)(19 15 67 23)(20 16 68 24), L1 = (96 48 49 1)(84 36 61 13)(72 24 73 25)(60 12 85 37)× ×(89 53 56 92)(90 65 55 80)(91 77 54 68)(66 67 79 78), L2 = (59 11 86 38)(71 23 74 26)(83 35 62 14)(95 47 50 2), F1 = (89 5 93 92)(77 17 81 80)(65 29 69 68)(53 41 57 86)× ×(1 4 40 37)(2 16 39 25)(3 28 38 13)(14 15 27 26), F2 = (73 6 81 91)(74 18 82 79)(75 30 83 67)(76 42 84 55), R1 = (40 88 9 57)(28 76 21 69)(16 64 33 81)(4 52 49 93)× ×(41 5 8 44)(42 17 7 32)(43 29 6 20)(18 19 31 30), R2 = (39 87 10 58)(27 75 22 70)(15 63 34 82)(3 51 46 94), B1 = (52 53 44 60)(51 65 32 59)(50 77 20 58)(49 89 8 57)× ×(9 12 48 45)(10 24 47 33)(11 36 46 21)(22 23 35 34), B2 = (54 72 43 64)(66 71 31 63)(78 70 19 62)(90 69 7 61), D1 = (57 60 96 93)(58 72 95 81)(59 84 94 69)× ×(70 71 83 82)(45 89 37 41)(46 90 38 42)(47 91 39 43)(48 92 40 44), D2 = (33 77 25 29)(34 78 26 30)(35 79 27 31)(36 80 28 32). Exercise 4.7.2 Verify that the properties of a permutation puzzle are satisfied for this puzzle.

4.8

Skewb

The skewb is a cube which has been subdivided into regions differently than the Rubik’s cube. First, fix an orientation of the cube in space, so we may talk about a front face, a back face, up, down, left, and right. Each of these 6 square faces are subdivided into 5 facets as follows:

4.8. SKEWB

75

The 4 corner facets are labeled exactly as in the case of the Rubik’s cube (as the lower case xyz, where x is the label of the face the facet lives on, y and z the two neighboring faces). The skewb itself is a cube subdivided as follows: there are 8 corner pieces which are each in the shape of a tetrahedron. For example, if you hold a cube in front of you the upper right hand corner of the front face is the facet of a tetrahedron whose facets are labels f1 , r4 , u2 . The moves of the skewb are different from the Rubik’s cube as well: Label the corners as XY Z, where xyz is the notation for any of the facets belonging to that corner piece. Pick a corner XY Z of the cube and draw a line L passing through that corner vertex and the opposite corner vertex (”skewering the cube”). That line defines a 120 degree rotation in the clockwise direction (viewed from the line looking down onto the corner you picked). One move of the skewb is defined in terms of this rotation as follows: Of course a 120 degree rotation of the entire cube about the line L will preserve the cube but swap some faces and some vertices. The skewb has a mechanism so that you can actually rotate half (a ”skewed” half) the skewb by 120 degrees about

76

CHAPTER 4. PERMUTATION PUZZLES

L and leave the other half completely fixed. This rotation of half the skewb about L will also be denoted XY Z. We may also label the 5 · 6 = 30 facets as follows:

+--------------+ | 20 17 | |

16

u

|

| 19 18 | +--------------+--------------+--------------+--------------+ | 5 2 | 10 7 | 25 22 | 30 27 | |

1

l

|

6

f

|

21

r

|

26

b

|

| 4 3 | 9 8 | 24 23 | 29 28 | +--------------+--------------+--------------+--------------+ | 15 12 | |

11

d

|

| 14 13 | +--------------+

Example 60 Consider the rotation U F R associated to the corner uf r. This move permutes the facets of the skewb. As a permutation, the disjoint cycle notation for this move is U F R = (6 16 21)(7 18 25)(10 17 24)(8 19 22). Note, in particular UFR does not move the 9-facet.

4.9. N × N RUBIK’S CUBE

77

The eight basic moves are given by F U R = (6 16 21)(7 18 25)(10 17 24)(8 19 22) RU B = (21 16 26)(22 17 30)(25 20 29)(23 18 27) BU L = (26 16 1)(27 20 5)(28 17 2)(30 19 4) LU F = (1 16 6)(2 19 10)(5 18 9)(3 20 7) F DR = (11 6 21)(25 13 9)(23 15 7)(24 12 8) BDR = (26 11 21)(29 13 23)(27 12 22)(30 14 24) F DL = (6 11 1)(9 15 3)(10 12 4)(8 14 2) LDB = (1 11 26)(3 13 27)(4 14 28)(5 15 29). All other moves are obtained by combining these moves sequentially. The reader who wishes may check these by xeroxing the above diagram, cutting it out and taping it together.

Exercise 4.8.1 Verify that the properties of a permutation puzzle are satisfied for this puzzle.

4.9

n × n Rubik’s cube

Other than the 2 × 2, 3 × 3, and 4 × 4 cubes, the only other Rubik’s cube manufactured, as far as I know, is the 5 × 5 Rubik’s cube. Apparently, there are mechanical problems which cause the manufacture of the n × n cubes to be overly expensive or perhaps even impossible, for n large. For information, at least theoretically, on the solution of such cubes, the reader might be interested in the article [L] or the postings in the archives of the ”cube-lovers” list [CL].

4.10

Pyraminx

The pyraminx is a puzzle in the shape of a tetrahedron. A tetrahedron is a 4-sided regular platonic solid, all of whose faces are equilateral triangles. Each of the 4 faces of the puzzle is divided into 9 triangular facets:

78

CHAPTER 4. PERMUTATION PUZZLES

There are a total of 4 · 9 = 36 facets on the pyraminx. They will be labeled as follows (the reader may want to xerox this, cut it out, then fold the corners and tape it into a tetrahedron):

4.10. PYRAMINX

79

We fix an orientation of the tetrahedron in space so that you are looking at a face which we call the ”front”. We may also speak of a ”right”, ”left”, and ”down” face. We label the 4 faces as f(ront), r(ight), l(eft), d(own). We label the vertices U(p), R(ight), L(eft), and B(ack). The tetrahedron itself has been subdivided into sub-tetrahedrons as follows: to each vertex X (so X ∈ {U, R, L, B}) there is an opposing face F of the solid. For each such face, we slice the solid along two planes parallel to the vertex X and lying in between the face and the vertex. We want these planes, along with the face and the vertex to be spaced apart equally. The sub-tetrahedrons in the slice of the face itself will be called the face slice associated to the face F , denoted F1 , the sub-tetrahedrons in the middle slice parallel to the face F will be called the middle slice associated to that face, denoted F2 , and the sub-tetrahedron containing the vertex X to the face tip associated to that vertex, denoted F3 . To each face labeled F , we have a clockwise rotation by 120 degrees of the first slice F1 of the face. We shall denote this rotation also by F1 . This rotation only moves the facets living on the slice F1 . Similarly, we have a

80

CHAPTER 4. PERMUTATION PUZZLES

clockwise rotation by 120 degrees of the second slice F2 of the face. We shall denote this rotation also by F2 . F3 denotes the clockwise rotation by 120 degrees of the opposing sub-tetrahedron containing the vertex X. These moves permute the labels for the 36 facets, hence may be regarded as a permutation of the numbers 1, 2, ..., 36. For example, the clockwise rotation by 120 degrees (looking at the front face) of the sub-tetrahedron opposite to the front face will be denoted F3 . The disjoint cycle notation for this move, regarded as a permutation, is F 3 = (23 22 36). The basic moves are given as follows: F1 = (2 32 27)(8 31 26)(7 30 12)(19 29 11)× ×(18 28 3)(1 17 13)(6 15 4)(5 16 14) F2 = (9 35 25)(21 34 24)(20 33 10) F3 = (23 22 36) R1 = (3 36 17)(11 34 16)(10 35 6)× ×(24 31 5)(23 32 1)(2 22 18)(9 20 7)(8 21 19) R2 = (12 33 15)(26 29 14)(25 30 4) R3 = (27 28 13) L1 = (1 28 22)(5 29 21)(4 33 9)× ×(14 34 8)(13 36 2)(3 27 23)(11 26 24)(12 25 10) L2 = (6 30 20)(16 31 19)(15 35 7) L3 = (17 32 18) D1 = (13 18 23)(14 19 24)(15 20 25)× ×(16 21 26)(17 22 27)(28 32 36)(29 31 34)(30 35 33) D2 = (4 7 10)(5 8 11)(6 9 12) D3 = (1 2 3) All other moves are obtained by combining these moves sequentially. Indeed, later, we shall want to use moves of the form F2 ∗ F3 , for each face F , but the disjoint cycle notation for these permutations are a little more cumbersome to write down. Exercise 4.10.1 Verify that the properties of a permutation puzzle are satisfied for this puzzle.

4.11. MEGAMINX

4.11

81

Megaminx

This puzzle is in the shape of a dodecahedron. A dodecahedron is a 12-sided regular platonic solid for which each of the 12 faces is a pentagon. We call two faces neighboring if they share an edge. There are 20 vertices and 30 edges on a dodecahedron. Each of the puzzle faces has been subdivided into 11 facets by slicing each edge with a cut which is both parallel to that edge and not far from the edge (say one-fifth the way to the opposite vertex). A picture is as follows:

There are a total of 11·12 = 132 facets on the puzzle. Each face of the solid is parallel to a face on the opposite side. Fix a face of the dodecahedron and consider a plane parallel to that face slicing through the solid and about onefifth the way to the opposite face. There are 12 such slices. Two such slices associated to two neighboring edges will intersect inside the dodecahedron at a 120 degree angle but two such slices associated to two non-neighboring edges will not intersect inside the dodecahedron (though they will intersect

82

CHAPTER 4. PERMUTATION PUZZLES

outside the solid of course). We slice up the solid dodecahedron in this way. This creates a smaller dodecahedron in the center and several other irregular smaller pieces. For each such slice associated to a given face fi there is a basic move still denoted fi of the megaminx given by clockwise rotating the slice of the megaminx by 120 degree, leaving the rest of the dodecahedron invariant. Such a move effects 26 facets of the megaminx and leaves the remaining 106 facets completely fixed. Label the 12 faces of the solid as f1 , f2 , ..., f12 in some fixed way. Imagine that the dodecahedron is placed in 3-space in such a way that one side on the xy-plane and is centered along the positive z-axis so that one of the vertices of the top face is at the xyz-coordinate (r, 0, s), where r is the radius of the inscribed circle for the pentagon and s is the distance from the ”up” face to the ”down” face of the dodecahedron.

Exercise 4.11.1 Suppose r = 1. Find s. (This is fairly hard - see the chapter on Platonic solids for some ideas.)

The up face we label as f1 . The others may be labeled according to the following graph, where faces are represented by vertices and two vertices are connected by an edge if the corresponding faces are neighboring.

4.11. MEGAMINX

83

A more symmetric way to order the faces of the dodecahedron is as follows (see [B], exercise 18.35): f1 f2 f3 f4 f5 f6 f7 f8 f9 f12 f11 f10

u u0 u1 u2 u3 u4 d2 d3 d4 d d1 d0

One property of this labeling is explained in the following

84

CHAPTER 4. PERMUTATION PUZZLES

Exercise 4.11.2 Suppose that the permutation (0 1 2 3 4) of the numbers {0, 1, 2, 3, 4} acts on the labels u0 , ..., u4 and d0 , ..., d4 in the obvious way. Show that this permutation of the faces corresponds to a rotation of the dodecahedron.

Notice that, like the cube, each vertex is uniquely determined by specifying the three faces it has in common. We use the notation x.y.z for the vertex of the dodecahedron which lies on the three faces x, y, z. Note that the order is irrelevent: x.y.z denotes the same vertex as y.x.z or z.y.x. The facets of the megaminx may be specified as with the Rubik’s cube: a corner facet may be specified as [x.y.z], where x is the face the facet lives on and y, z are the two neighboring faces of the facet. An edge facet may be specified by [x.y], where x is the face the facet lives on and y is the other neighboring face of the facet. The center facet of f1 will simply be denoted by [f1 ]. We will call this label the intrinsic label. We may label the facets of the up face f1 as follows:

f1 facet symbol a b c d e f g h i j k

numerical label 1 2 3 4 5 6 7 8 9 10 11

intrinsic label [f1 .f6 .f2 ] [f1 .f2 ] [f1 .f2 .f3 ] [f1 .f3 ] [f1 .f3 .f4 ] [f1 .f4 ] [f1 .f4 .f5 ] [f1 .f5 ] [f1 .f5 .f6 ] [f1 .f6 ] [f1 ]

For the next face (the f2 face), we label the facets in such a way that the abc edge of f1 joins the ghi edge of f2 :

4.11. MEGAMINX f2 facet symbol a b c d e f g h i j k

85 numerical label 12 13 14 15 16 17 18 19 20 21 22

intrinsic label [f2 .f6 .f7 ] [f2 .f7 ] [f2 .f7 .f8 ] [f2 .f8 ] [f2 .f8 .f3 ] [f2 .f3 ] [f2 .f3 .f1 ] [f2 .f1 ] [f1 .f5 .f6 ] [f2 .f6 ] [f2 ]

In general, we can label the remaining facets in such a way that the

86

CHAPTER 4. PERMUTATION PUZZLES

basic moves are, as permutations, given by: f1 = (1 3 5 7 9)(2 4 6 8 10)(20 31 42 53 64)× ×(19 30 41 52 63)(18 29 40 51 62) f2 = (12 14 16 18 20)(13 15 17 19 21)(1 60 73 84 31)× ×(3 62 75 86 23)(2 61 74 85 32) f3 = (23 25 27 29 31)(24 26 28 30 32)(82 95 42 3 16)× ×(83 96 43 4 17)(84 97 34 5 18) f4 = (34 36 38 40 42)(35 37 39 41 43)(27 93 106 53 5)× ×(28 94 107 54 6)(29 95 108 45 7) f5 = (45 47 49 51 53)(46 48 50 52 54)(38 104 117 64 7)× ×(39 105 118 65 8)(40 106 119 56 9) f6 = (56 58 60 62 64)(57 59 61 63 65)(49 115 75 20 9)× ×(50 116 76 21 10)(51 117 67 12 1) f7 = (67 69 71 73 75)(68 70 72 74 76)(58 113 126 86 12)× ×(59 114 127 7 13)(60 115 128 78 14) f8 = (78 80 82 84 86)(79 81 83 85 87)(71 124 97 23 14)× ×(72 125 98 24 15)(73 126 89 25 16) f9 = (89 91 93 95 97)(90 92 94 96 98)(80 122 108 34 25)× ×(81 123 109 35 26)(82 124 100 36 27) f10 = (100 102 104 106 108)(101 103 105 107 109)× ×(91 130 119 45 36)(92 131 120 46 37)(93 122 111 47 38) f11 = (111 113 115 117 119)(112 114 116 118 120)× ×(102 128 67 56 47)(103 129 68 57 48)× ×(104 130 69 58 49) f12 = (122 124 126 128 130)(123 125 127 129 131)× ×(100 89 78 69 111)(101 90 79 70 112)(102 91 80 71 113)

4.12

Other permutation puzzles

I have left out several puzzles: ”topspin” and ”turnstile” (planar puzzles), ”mozaika” (an equator-like puzzle, but the hemispheres may be rotated), ”alexander’s star” (a stellated icosahedron), ”the ”impossiball” (a spherically shaped icosahedron - see [H]), ”mickey’s challenge” (a spherically shaped irregular polyhedron - essentially the same as the skewb but with some added orientations of faces). The puzzle ”Christoph’s jewel”, essentially a ”Rubik octahedron”, may be solved using ”super-Rubik’s cube moves” (see [H]). (Indeed, one may take

4.12. OTHER PERMUTATION PUZZLES

87

a Rubik’s cube, strip off all the stickers (using soap and water), and replace them with new stickers modeling a Rubik octahedron. This is because the octahedron is the dual solid of the cube, as described in chapter 7 below ”Symmetry groups of the Platonic solids”.) The ”orbix” puzzle (a battery run puzzle which has 12 buttons which light up) is a permutation puzzle if you think of a move (which switches certain of the buttons on/off) as permuting the elements of the set of all subsets of the 12 buttons (the subset of buttons which are lit) amongst themselves. There is some mention of such puzzles in, for example, [Si], [H], [B] and [GT].

88

CHAPTER 4. PERMUTATION PUZZLES

Chapter 5 Groups, I Q: ”What’s commutative and purple?” A: ”An abelian grape”. - Ancient Math Joke ”In 1910 the mathematician Oswald Veblen and the physicist James Jeans were discussing the reform of the mathematical curriculum at Princeton University. ‘We may as well cut out group theory,‘ said Jeans. ‘That is a subject which will never be of any use to physics.‘ It is not recorded whether Veblen disputed Jeans’ point, or whether he argued for the retention of group theory on purely mathematical grounds. All we know is that group theory continued to be taught. And Veblen’s disregard for Jeans’ advice continued to be of some importance to the history of science at Princeton. By the irony of fate group theory later grew into one of the central themes of physics, and it still dominates the thinking of all of us who are struggling to understand the fundamental particles of nature.” Freeman J. Dyson SCIENTIFIC AMERICAN, Sep, 1964 When we studied permuation puzzles in Chapter 4, recall that one of the criteria was that each move was ”invertible”. This is, in fact, one of the conditions for the set of all legal moves of a permutation puzzle to form a group. A group is a set G with a binary operation (namely a function ∗ : G × G →) satisfying certain properties to be given later, one of which is 89

90

CHAPTER 5. GROUPS, I

that each element has an inverse element associated to it. One should be a little careful, since not every permutation puzzle gives rise to a group in this way. For example, the set of moves of the 15 puzzle do not form a group in this way though the set of moves of the Rubik’s cube group do. Just as for sets, we must decide on how to describe a group. If G is finite then one way is to list all the elements in G and list (or tabulate) all the values of the function ∗. Another method is to describe G in terms of some properties and then define a binary operation ∗ on G. A third method is to give a ”presentation” of G. Each of these has its advantages and disadvantages. We shall eventually introduce all three of these approaches. First, we start with an example.

5.1

The symmetric group

Before defining anything, we shall provide a little motivation for some general notions which will arise later. Let X be any finite set and let SX denote the set of all permutations of X onto itself: SX = {f : X → X | f is a bijection}. This set has the following properties: 1. if f, g belong to SX then f g (the composition of these permutations) also belongs to SX , (”closed under compositions”), 2. if f, g, h all belong to SX then (f g)h = f (gh), (”associativity”), 3. the identity permutation I : X → X belongs to SX (”existence of the identity”), 4. if f belongs to SX then the inverse permutation f −1 also belongs to SX (”existence of the inverse”). The set SX is called the symmetric group of X. We shall usually take for the set X a set of the form {1, 2, ..., n}, in which case we shall denote the symmetric group by Sn . This group is also called the symmetric group on n letters. Example 61 : Suppose X = {1, 2, 3}. We can describe SX as SX = {I, s1 = (1 2), s2 = (2 3), s3 = (1 3 2), s4 = (1 2 3), s5 = (1 3)}.

5.2. GENERAL DEFINITIONS

91

We can compute all possible products of two elements of the group and tabulate them in a multiplication table is

I s1 s2 s3 s4 s5

I I s1 s2 s3 s4 s5

s1 s1 I s4 s5 s2 s3

s2 s2 s3 I s1 s5 s4

s3 s3 s2 s5 s4 I s1

s4 s4 s5 s1 I s3 s2

s5 s5 s4 s3 s2 s1 I

Exercise 5.1.1 Verify the four properties of SX mentioned above. (Note that the verification of associativity follows from the associative property of the composition of functions - see the Exercise 3.0.4).

5.2

General definitions

We take the above four properties of the symmetric group as the four defining properties of a group: Definition 62 Let G be a set and suppose that there is a mapping ∗:G×G×G (g1 , g2 ) 7−→ g1 ∗ g2 (called the group’s operation) satisfying (G1) if g1 , g2 belong to G thn g1 ∗ g2 belongs to G (”G is closed under ∗”), (G2) if g1 , g2 , g3 belong to G then (g1 ∗g2 )∗g3 = g1 ∗(g2 ∗g3 ) (”associativity”), (G3) there is an element 1 ∈ G such that 1 ∗ g = g ∗ 1 = g for all g ∈ G (”existence of an identity”), (G4) if g belongs to G then there is an element g −1 ∈ G, called the inverse of g such that g ∗ g −1 = g −1 ∗ g = 1 (”existence of inverse”). Then G (along with the operation ∗) is a group.

92

CHAPTER 5. GROUPS, I

Example 63 Actually, this is a ”non-example”. Let S be the set of all legal moves (one can eventually make from a legally obtained position) of the 15 puzzle (as described in Chapter 4). In a given position, for example the solved position, there aren’t that many possibilities: there are only 2 moves in the solved position and there are never any more that 4 moves possible from any position. From the solved position one can move (15, 16) and (12, 16) (where 16 denotes the blank square) but not for example (1, 16). Since (15, 16), (1, 16) ∈ S and since (1, 16)(15, 16) is not a legal move, it follows that composition of legal moves is not always legal. This shows that composition is not a binary operation, so property number (G1) fails to hold. In the above definition, we have not assumed that there was exactly one identity element 1 of G because, in fact, one can show that if there is one then it is unique. (To do this you need to use the cancellation law: if a ∗ c = b ∗ c, where a, b, c ∈ G, then a = b.) Likewise, if G is a group and g ∈ G then the inverse element of g is unique. There are other properties of a group which can be derived from (G1)-(G4). We shall prove them as needed. The multiplication table of a finite group G is a tabulation of the values of the binary operation ∗. Let G = {g1 , ..., gn }. The multiplication table of G is: * g1 g2 .. . gi .. .

g1

g2

...

gj

... gn

gi ∗ gj

gn Some properties: Lemma 64 (a) Each element gk ∈ G occurs exactly once in each row of the table. (b) Each element gk ∈ G occurs exactly once in each column of the table. (c) If the (i, j)th entry of the table is equal to the (j, i)th entry then gi ∗gj = gj ∗ gi . (d) If the table is symmetric about the diagonal then g ∗ h = h ∗ g for all g, h ∈ G. (In this case, we call G abelian.)

5.2. GENERAL DEFINITIONS

93

Example 65 Let C12 be the group whose elements are {0, 1, ..., 11} and for which the group operation is simply ”addition mod 12”, just as one adds time on a clock (except that we call ”12 o’clock” ”0 o’clock”). Thus 5 + 8 = 1, 1 + 11 = 0, and so on. Question: What is the inverse element of 5? The inverse of 1? This group is called the cyclic group of order 12. Exercise 5.2.1 Compute the multiplication table for C12 . Definition 66 Let n > 1 be an integer and let Cn be the group whose elements are {0, 1, ..., n − 1} (more precisely, {0, 1, ..., n − 1}, where i is the residue class mod n of i) and for which the group operation is simply ”addition mod n”. This group is called the cyclic group of order n. For further details on cyclic groups, see for example [G] or [R]. Definition 67 Let g and h be two elements of a group G. We say that g commutes with h (or that g, h commute) if g ∗ h = h ∗ g. We call a group commutative (or ”abelian”) if every pair of elements g, h belonging to G commute. If G is a group which is not necessarily commutative then we call G noncommutative (or ”nonabelian”). Example 68 The integers, with ordinary addition as the group operation, is an abelian group. Exercise 5.2.2 Show that any group having exactly 2 elements is abelian. Now the reader should understand the punchline to the joke quoted at the beginning! Convention: When dealing with groups in general we often drop the ∗ and denote multiplication simply by juxtaposition (that is, sometimes we write gh in place of g ∗ h), with one exception. If the group G is abelian then one often replaces ∗ by + and then + is not dropped. Now that we know the definition of a group, the question arises: how might they be described? The simplest answer is that we describe a group much as we might describe a set: we could list all its elements and give the multiplication table or we could describe all its elements and their multiplication in terms of some property from which we can verify the four properties of group. Though the first way has the distinct advantage of being explicit, it

94

CHAPTER 5. GROUPS, I

is this second alternative which is the most common since it is usually more concise. Our objective is to introduce terminology and techniques which enable us to analyse mathematically permutation puzzles. The type of groups which arise in this context are defined next. Definition 69 Let X be a finite set. Let g1 , g2 , ..., gn be a finite set of elements of permutations of X (so that they all belong to SX ). Let G be the set of all possible products of the form g = x1 ∗ x2 ... ∗ xm ,

m > 0,

where each of the x1 , ..., xm is taken from the set {g1 , ..., gn }. The set G, together with the group operation given by composition of permutations, is called a permutation group with generators g1 , ..., gn . We sometimes write G =< g1 , ..., gn >⊂ SX . It is not too hard to justify our terminology: Lemma 70 A permutation group is a group. proof: Let G be a permutation group as in the above definition. We shall only prove that each g ∈ G has an inverse, leaving the remainder of the properties for the reader to verify. The set {g n | n ≥ 1} ⊂ SX is finite. There are n1 > 0, n2 > n1 such that n1 g = g n2 . Then g −1 = g n2 −n1 −1 since g · g n2 −n1 −1 = 1. 2 Remark 3 The above definition can be generalized: Replace SX by any group S which includes all the generators g1 , ..., gn . The resulting set G is called the group generated by the elements g1 , ..., gn . Algorithm: Input: The generators g1 , ..., gn (as permutations in SX ), Output: The elements of G, S = {g1 , ..., gn , g1−1 , ..., gn−1 }, L = S ∪ {1},

5.2. GENERAL DEFINITIONS

95

for g in S do for h in L do if g*h not in L then L = L union {g*h} endif endfor endfor Note that the size of the list L in the for loop changes after each iteration of the loop. The meaning of this is that the if-then command is to be executed exactly once for each element of the list. Exercise 5.2.3 Verify that permutation group G satisfies the four properties of a group (G1)-(G4). Definition 71 If G is a group then the order of G, denoted |G|, is the number of elements of G. If g is an element of the group G then the order of g, denoted ord(g), is the smallest positive integer m such that g m = 1, if it exists. If such an integer m does not exist then we say that g has ”infinite order”. Example 72 For example, there is an even permutation of order 42 in S12 , for example (1, 2)(3, 4, 5)(6, 7, 8, 9, 10, 11, 12), and an odd permutation of order 15 in S8 , for example (1, 2, 3)(4, 5, 6, 7, 8). Singmaster [Si] states that the maximal order in the Rubik’s cube group is 1260. We shall be able to make use of the following fact frequently. Theorem 73 (a) (Cauchy) Let p be a prime dividing |G|. There is a g ∈ G of order p. (b) (Lagrange) Let n be an integer not dividing |G|. There does not exist a g ∈ G of order n. This will be proven a little later. As an application of this: we shall see later that the Rubik’s cube group G has the property that |G| = 227 314 53 72 11. It follows from this and Lagrange’s theorem that there is no move of the Rubik’s cube of order 13 but there is one of order 11.

96

CHAPTER 5. GROUPS, I

Exercise 5.2.4 Let X = {1, 2, 3}. We use the notation of the example above. (a) Let G be the permutation group with generator s1 , G =< s1 >. Verify that there are only two elements in G. (b) What is the order of s5 ? (c) Let G be the permutation group with generator s3 , G =< s3 >. Verify that there are only three elements in G. (d) Find the order of s3 . (e) Show that SX =< s1 , s2 >. Definition 74 If G is a permutation group G with only one generator then we say that G is cyclic. Lemma 75 If G =< g > is cyclic with generator g then |G| = ord(g). proof: Let m = ord(g), so g m = 1. We can list all the elements of G as follows: 1, g, g 2 , ..., g m−1 . There are m elements in this list. 2

5.2.1

The Gordon game

Let G be a finite group, written G = {g0 = 1, g1 , ..., gn }. You and your opponent share a set of move tokens, denoted M = {g1 , ..., gn }, and place tokens, denoted P = {g1 , ..., gn }. Rules to play: • Players alternate turns. Each turn consists of removing one move token and one place token according to the conditions listed below. The first person who cannot make a legal play loses. Let m0 = p0 = 1 and let i = 1.

5.2. GENERAL DEFINITIONS

97

• First player picks any move token m1 ∈ M and the place token p1 = m1 ∈ P . These tokens m1 and p1 are then removed from M and P , resp.. • The next player picks any move token mi+1 such that pi+1 = mi+1 pi ∈ P . These tokens mi+1 and pi+1 are then removed from M and P , resp.. • Increment i and go to the previous step. Example 76 Let G = F7 = {0, 1, 2, 3, 4, 5, 6}. The moves of a game are determined by recording the move tokens. One possible game is • 4 1 3 2 0 1 2 3 4 5 6 where the • over the identity element 0 of the group indicates that it isn’t moved and the numbers above a group element indicates when it was moved: 1st : m1 = 2, p1 = 2; 2nd : m2 = 4, p2 = 6; 1st : m3 = 6, p3 = 5; 2nd : m4 = 3, p4 = 1; 2nd player wins Exercise 5.2.5 Play a game! Remark 4 If G = Z/pZ (the cyclic group with p elements) there is a conjecture that the 2nd player has a winning strategy when p > 5 (see Isbell’s note [I]). In general, strategies are not not only not known, they haven’t even been conjectured. Remark 5 If you and your opponent both try to drag the game on as long as possible, can you exhaust the set of move tokens and the set of place tokens? The answer is known for abelian groups, dihedral groups and groups of order < 32. The general answer is unknown.

98

CHAPTER 5. GROUPS, I

5.3

Subgroups

Definition 77 Let G be a group. A subgroup of G is a subset H of G such that H, together with the operation ∗ inherited as a subset of G, satisfies the group operations (G1)-(G4) (with G replaced by H everywhere). Notation: If G is a group then we will denote the statement ”H is a subgroup of G” by H < G. Problem: What are the subgroups of the Rubik’s cube group? It turns out that there are too many to list but later, when we have a more useful way of describing a group (using generators and relations - see §9.3), we will explicitly determine some of the subgroups of ”small” order. Theorem 78 (Lagrange) Let H be a subgroup of a finite group G. Then |H| divides |G|. proof: For x, y ∈ G, define x ∼ y if xH = yH, where xH = {x ∗ h | h ∈ H}. This is an equivalence relation (Exercise: Check reflexive, symmetry, and transitivity). Moreover, the equivalence class of x consists of all elements in G of the form x ∗ h, for some h ∈ H, i.e., [x] = xH. Let g1 , ..., gm ∈ G denote a complete set of representatives for the equivalence classes of G. Because of the cancellation law for groups, |xH| = |H| for each x ∈ G. Furthermore, we know that the equivalence classes partition G, so m G = ∪m i=1 [gi ] = ∪i=1 gi H.

Comparing cardinalities of both sides, we obtain |G| = |g1 H| + ... + |gm H| = m|H|. This proves the theorem. 2 Definition 79 If H and G are finite groups and H < G then the integer |G|/|H| is called the index of H in G, denoted [G : H] = |G|/|H|. Exercise 5.3.1 Show, as a corollary to the previous Theorem 78, that Theorem 73 is true.

5.3. SUBGROUPS

99

Example 80 A permutation group G generated by elements g1 , ..., gn belonging to SX is a subgroup of SX , i.e., G < SX . Example 81 Let AX = {g ∈ SX | g is even}. This is a subgroup of Sn called the alternating subgroup of degree n . Example 82 The collection of all moves of the 15 puzzle may be viewed as a subgroup of S16 . Definition 83 The center of a group G is the subgroup Z(G) of all elements which commute with every element of G: Z(G) = {z ∈ G | z ∗ g = g ∗ z, for all g ∈ G}. Of course, the identity element always belongs to G. If the identity element is the only element of Z(G) then we say G has trivial center. On the other hand, G is commutative if and only if G = Z(G). Exercise 5.3.2 Let G = S3 . Determine Z(G). Example 84 The collection of all moves of the Rubik’s cube may be viewed as a subgroup G of S48 . The center of G consists of exactly two elements, the identity and the ”superflip” move which has the effect of flipping over every edge, leaving all the corners alone and leaving all the subcubes in their original position. One move for the superflip is superflip = R ∗ L ∗ F ∗ B ∗ U ∗ D ∗ R ∗ L ∗ F ∗ B ∗ U ∗ F 2 ∗ MR ∗ ∗F 2 ∗ U −1 ∗ MR2 ∗ B 2 ∗ MR−1 ∗ B 2 ∗ U ∗ MR2 ∗ D, where MR is middle right slice rotation by 90 degrees (viewed from the right face). The proof of this fact uses the determination of the group structure of G given later (see also [B]).

100

5.4

CHAPTER 5. GROUPS, I

Example: The dihedral group

Pick an integer n > 2 and let R be a regular n-gon centered about the origin in the plane. If n = 3 then R is an equilateral triangle, if n = 4 then R is a square, if n = 5 then R is a pentagon, and so on. Let G denote the set of all linear transformations of the plane to itself which preserve the figure R. The binary operation ◦ : G × G → G given by composition of functions gives G the structure of a group. This group is called the group of symmetries of R. Label the vertices of the n-gon as 1, 2, ..., n. The group G permutes these vertices amongst themselves, hence each g ∈ G may be regarded as a permutation of the set of vertices V = {1, 2, ..., n}. In this way, we may regard G as a permutation group since it is the subgroup of Sn generated by the elements of G. The fact that this group has 2n elements follows from a simple counting argument: Let r ∈ G denote the element which rotates R by 2π/n radians counterclockwise about the center. Let L be a line of symmetry of R which bisects the figure into two halves. Let s denote the element of G which is reflection about L. There are n rotations by a multiple of 2π/n radians about the center in G: 1, r, r2 , ..., rn−1 . There are n elements of G which are composed of a reflection about L and a rotations by a multiple of 2π/n radians about the center: s, s ◦ r, s ◦ r2 , ..., s ◦ rn−1 . These comprise all the elements of G. One remarkable property of this symmetry group, which we shall use in the example in the next section, is that it is generated by any two distinct reflections in the group: Lemma 85 Pick two distinct lines L, L0 of symmetries of R, each of which bisects R in half, and let s, s0 (resp.) denote the corresponding reflections, regarded as elements in Sn . Then G =< s, s0 >. The interested reader is referred to [NST], [R], or [Ar], chapter 5, §3, for a proof. The symmetry group of R is closely related to a group known as the dihedral group of order 2n, denoted D2n . We shall state the precise relation in a later chapter (chapter 8) after we have introduced more terminology. Example 86 Let G be the symmetry group of the square: i.e., the group of

5.5. EXAMPLE: THE TWO SQUARES GROUP symmetries of the square generated by the rigid motions g0 = 90 degrees clockwise rotation about O, g1 = reflection about `1 , g2 = reflection about `2 , g3 = reflection about `3 , g4 = reflection about `4 , where `1 , `2 , `3 denote the lines of symmetry in the picture below:

The elements of G are 1, g0 , g02 , g03 , g1 , g2 , g3 , g4 . Let X be the set of vertices of the square. Then G acts on X.

5.5

Example: The two squares group

This material is based on an idea mentioned in [FS].

101

102

CHAPTER 5. GROUPS, I

Let H =< R2 , U 2 > denote the group generated by the two square moves, R2 and U 2 or the Rubik’s cube. (The reader with a cube in hand may want to try the Singmaster magic grip : the thumb and forefinger of the right hand are placed on the front and back face of the fr, br edge, the thumb and forefinger of the left hand are placed on the front and back face of the uf, ub edge; all moves in this group can be made without taking your fingers off the cube.) This group contains the useful 2-pair edge swap move (R2 ∗ U 2 )3 .

We can find all the elements in this group fairly easily:

H = {1, R2 , R2 ∗ U 2 , R2 ∗ U 2 ∗ R2 , (R2 ∗ U 2 )2 , (R2 ∗ U 2 )2 ∗ R2 , (R2 ∗ U 2 )3 , (R2 ∗ U 2 )3 ∗ R2 , (R2 ∗ U 2 )4 , (R2 ∗ U 2 )4 ∗ R2 , (R2 ∗ U 2 )5 , (R2 ∗ U 2 )5 ∗ R2 },

Therefore, |H| = 12. Note that 1 = (R2 ∗ U 2 )6 , U 2 = (R2 ∗ U 2 )5 ∗ R2 , and U 2 ∗ R2 = (R2 ∗ U 2 )5 . (By the way, this listing without repetition of H by successive multiplication by R2 then U 2 may be reformulated graphically by saying the ”the Cayley graph of H with generators R2 , U 2 has a Hamiltonian circuit”. This interpretation will be discussed in the next chapter.)

To discover more about this group, we label the vertices of the cube as follows:

5.5. EXAMPLE: THE TWO SQUARES GROUP

103

The move R2 acts on the set of vertices by the permutation (1 4)(2 3) and the move U 2 acts on the set of vertices by the permutation (4 5)(3 6). We label the vertices of a hexagon as follows:

104

CHAPTER 5. GROUPS, I

hexagon The permutation (1 4)(2 3) is simply the reflection about the line of symmetry containing both 5 and 6. The permutation (4 5)(3 6) is simply the reflection about the line of symmetry containing both 1 and 2. By a fact stated in section 5.4, these two reflections generate the symmetry group of the hexagon.

5.6

Commutators

Definition 87 If g, h are two elements of a group G then we call the element [g, h] = g ∗ h ∗ g −1 ∗ h−1 then commutator of g, h. Not that [g, h] = 1 if and only if g, h commute. Thus the commutator may be regarded as a rough measurement of the lack of commutativity.

5.6. COMMUTATORS

105

Exercise 5.6.1 Let G = S3 , the symmetric group on 3 letters. Compute the commutators [s1 , s2 ], [s2 , s1 ]. Exercise 5.6.2 Let R, U be as in the notation for the Rubik’s cube moves introduced in the previous chapter. Determine the order of the move [R, U ]. (Ans: 6) Definition 88 (Singmaster [Si]) Let G be the permutation group generated by the permutations R, L, U, D, F, B regarded as permutations in S54 . The Y commutator is the element [F, R−1 ] = F ∗ R−1 ∗ F −1 ∗ R. The Z commutator is the element [F, R] = F ∗ R ∗ F −1 ∗ R−1 . Exercise 5.6.3 (a) Find the orders of the Y commutator and the Z commutator. (b) Find the order of [R, [F, U ]]. Example 89 If x, y are basic moves of the Rubik’s cube associated to faces which share an edge then (a) [x, y]2 permutes exactly 3 edges and does not permute any corners, (b) [x, y]3 permutes exactly 2 pairs of corners and does not permute any edges. Definition 90 Let G be any group. The group G0 generated by all the commutators {[g, h] | g, h belong to G} This is called the commutator subgroup of G. This group may be regarded as a rough measurement of the lack of commutativity of the group G. Remark 6 We will see later that the group generated by the basic moves of the Rubik’s cube - R, L, U, D, F, B - has a relatively large commutator subgroup. In other words, roughly speaking ”most” moves of the Rubik’s cube can be generated by commutators such as the Y commutator or the Z commutator.

106

5.7

CHAPTER 5. GROUPS, I

Conjugation

Definition 91 : If g, h are two elements of a group G then we call the element g h = h−1 ∗ g ∗ h the conjugation of g by h. Note that g h = 1 if and only if g, h commute. Thus the conjugates may be regarded as a rough measurement of the lack of commutativity. Exercise 5.7.1 Show g ∗ [g −1 , h−1 ] = g h . Exercise 5.7.2 Let G = S3 , the symmetric group on 3 letters, in the notation of the example above. Compute the conjugations ss12 ,

ss21 .

Exercise 5.7.3 Let R, U be as in the notation for the Rubik’s cube moves introduced in the previous chapter. Determine the order of the move RU . (Ans: 4) Definition 92 : We say two elements g1 , g2 of G are conjugate if there is an element h ∈ G such that g2 = g1h . It turns out that it is easy to see when two permutations g, h ∈ Sn are conjugate: they are conjugate if and only if the cycles in their respective disjoint cycle decompositions have the same length when assanged from shortest to longest. For example, the elements g = (6, 9)(1, 3, 4)(2, 5, 7, 8),

h = (1, 2)(3, 4, 5)(6, 7, 8, 9)

are conjugate. We shall leave the details and the proof for later - see §8.3.1 Exercise 5.7.4 Show that the notion of conjugate defines an equivalence relation. That is, show that (a) any element g ∈ G is conjugate to itself (”reflexive”), (b) if g is conjugate to h (g, h belonging to G) then h is conjugate to g (”symmetry”), (c) if g1 is conjugate to g2 and g2 is conjugate to g3 then g1 is conjugate to g3 (”transitivity”).

5.7. CONJUGATION

107

Notation: The set of equivalence classes of G under the equivalence relation given by conjugation, will be denoted G∗ . The polynomial X pG (t) = tord(g) , g∈G∗

is called the generating polynomial of the order function on G. Note two elements which are conjugate must have the same order since (h−1 gh)n = (h−1 gh)(h−1 gh)...(h−1 gh) = h−1 g n h, for n = 1, 2, ... and g, h ∈ G. In [Si], §5.10D, D. Singmaster asks for the possible orders of the elements of the Rubik’s cube group and how many elements of each order there are. (A method for determining this will be described later in this text.) This question of Singmaster motivates the following: Problem: Determine pG (t) for the Rubik’s cube group. Example 93 For S8 , the generating polynomial is t + 4t2 + 2t3 + 4t4 + t5 + 5t6 + t7 + t8 + t10 + t12 + t15 and for S12 it is t + 6t2 + 4t3 + 9t4 + 2t5 + 16t6 + t7 + 4t8 + 2t9 + 6t10 + t11 + 9t12 + 2t14 + 2t15 + t18 + 2t20 + t21 + t24 + t28 + 3t30 + t35 + t42 + t60 . (Both of these calculations were performed by MAPLE.) For example, it follows that there is an even permutation of order 42 in S12 and an odd permutation of order 15 in S8 . Singmaster [Si] states that the maximal order in the Rubik’s cube group is 1260. Definition 94 : Fix an element g in a group G. The set Cl(g) = {h−1 ∗ g ∗ h | h ∈ G} is called the conjugacy class of g in G. It is the equivalence class of the element g under the relation given by conjugation. If H is a subgroup of G and if g is a fixed element of G then the set H g = {g −1 ∗ h ∗ g | h ∈ H} is a subgroup of G. Such a subgroup of G is called a subgroup conjugate to H.

108

CHAPTER 5. GROUPS, I

Exercise 5.7.5 Let S be the set of all subgroups of G. We define a relation R on S by R = {(H1 , H2 ) ∈ S × S | H1 is conjugate to H2 }. Show that R is an equivalence relation. Exercise 5.7.6 Let G = Sn and let H =< g > be a cyclic subgroup generated by a permutation g of the set {1, 2, ..., n}. With respect to the equivalence relation in the previous exercise, show that a subgroup K of G belongs to the equivalence class [H] of H in G if any only if K is cyclic and is generated by an element k of G conjugate to g ∈ G.

5.8

Orbits and actions

Definition 95 Let X be a set and let G be a group. We call X a G-set and we say G acts on X provided the following conditions hold: 1. each g belonging to G gives rise to a function φg : X → X, 2. the identity 1 of the group G defines the identity function on X, 3. if g, h belong to G then the composite φgh : X → X satisfies φgh (x) = φh (φg (x)). We call this action a left action since the left-most element (namely, g) in the product gh acts first. Similarly, we define Definition 96 Let X be a set and let G be a group. We say G acts on X on the right provided the following conditions hold: 1. each g belonging to G gives rise to a function φg : X → X,

5.8. ORBITS AND ACTIONS

109

2. the identity 1 of the group G defines the identity function on X, 3. if g, h belong to G then the composite φgh : X → X satisfies φgh (x) = φg (φh (x)). We call this action a right action since the right-most element (namely, h) in the product gh acts first. Remark 7 (1) We shall see another interpretation of these definitions in the later chapter entitled, ”Groups, II”. (2) Given a left action φg , one can create a right action by defining φ0g = φg−1 . Following the standard convention, the Rubik’s cube will act on the set of facets of the cube on the right. Definition 97 Let G act on a set X. We call the action transitive if for each pair x, y belonging to X there is a g ∈ G such that y = φg (x). In other words, a group G acts transitively on a set X if any element x of X can be send to any other element y of X by some element of G. Example 98 Let X be a finite set and let G = SX be the symmetric group of X. Then X is a G-set and G acts transitively on X. Exercise 5.8.1 Show that the action in the previous example is transitive. Example 99 Let G be the group of all 2 × 2 invertible matrices with real entries, G = GL2 (R). This group acts on the set of column vectors on the left. Exercise 5.8.2 Let G be the permutation group generated by the permutations R, L, U, D, F, B, regarded as elements of S48 . Let E denote the set of edges of the cube, which we identify with the set of edge subcubes. Let V denote the set of vertices of the cube, which we identify with the set of corner subcubes of the cube. Let X be the set of all movable subcubes of the Rubik’s cube (which may identify as the union of E and V ). Then G acts on X, E and G acts on V . Question (a) Is the action of G on X transitive? (a) Is the action of G on E transitive? (b) Is the action of G on V transitive?

110

CHAPTER 5. GROUPS, I

Exercise 5.8.3 Let G be a group and let X = G. Define left multiplication of G on X by: φg : X → X x 7−→ φg (x) = g ∗ x. (a) Show that left multiplication defines a left action of G on X. (b) Show that this action is transitive. (c) Show that each φg : G → G is a permutation of the set G, so φg ∈ SG . Exercise 5.8.4 Let G be a group and let X = G. Define right multiplication of G on X by: φg : X → X x 7−→ φg (x) = x ∗ g. (a) Show that right multiplication defines a right action of G on X. (b) Show that this action is transitive. (c) Show that each φg : G → G is a permutation of the set G, so φg ∈ SG . Exercise 5.8.5 Let G be a group and let X = G. Define conjugation on X by: φg : X → X x 7−→ φg (x) = g −1 ∗ x ∗ g. Show that conjugation defines an action of G on X (X and G as above). Exercise 5.8.6 Let G be a group and let X denote the set of all subgroups of G. Define conjugation on X by: φg : X → X x 7−→ φg (x) = g −1 ∗ x ∗ g. Show that this defines an action of G on X. Remark 8 In general, the actions in the last two exercises are not transitive. Definition 100 Let G be a group acting on a set X. For each x belonging to X, the set G ∗ x = {φg (x) | g ∈ G} is called the orbit of x ∈ X under G.

5.8. ORBITS AND ACTIONS

111

Algorithm Input: A set S of generators of a permutation group G and an x belonging to X Output: The orbit of x, G ∗ x orbit = {x} for y in orbit do for g in S do if g*y not in orbit then orbit = orbit union {g*y} endif endfor endfor Note that the size of the list orbit in the for loop changes after each iteration of the loop. As mentioned before, the meaning of this is that the if-then command is to be executed exactly once for each element of the list. Exercise 5.8.7 Let G be the Rubik’s cube group and let x be the uf edge facet. Find the orbit of x under the action of G using the above algorithm. Show each step. Exercise 5.8.8 Let G be the group of moves of the Rubik’s cube and let X be the set of vertices of the cube. Let H be the subgroup of G generated by U ∗ R. Find: (1) the order of U ∗ R, (Ans: 105) (2) the orbit (in the Singmaster notation) of the uf r vertex in X under H. Definition 101 Let G be a group acting on a set X with the action denoted by φ. For each x belonging to X, the subgroup stabG (x) = Gx = {g ∈ G | φg (x) = x} is called the stabilizer of x in G. Exercise 5.8.9 Let G be a group acting on a set X, φg : X → X, for all g ∈ G. Show that, for all x ∈ X and all g ∈ G, we have stabG (φg (x)) = g ∗ stabG (x) ∗ g −1 . Example 102 Let G be the group of symmetries of the square (see the example above), let X be the set of vertices of the square, and let x0 be the vertex in the lower right hand corner. Then stabG (x0 ) =< g3 >.

112

CHAPTER 5. GROUPS, I

Exercise 5.8.10 Let G be any group and let X = G. Let G act on X by left multiplication: φg : X → X x 7−→ φg (x) = g ∗ x. Show that stabG (x) = 1, for all x belonging to X = G. Exercise 5.8.11 Let G be any group and let X = G. Let G act on X by conjugation: φg : X → X x 7−→ φg (x) = g ∗ x ∗ g −1 . Show that stabG (x) = {g ∈ G | g ∗ x = x ∗ g}, for all x belonging to X = G. (The subgroup CG (x) = {g ∈ G | g ∗ x = x ∗ g} is called the centralizer of x in G.) Example 103 Let X be the set of consisting of the 48 facets of the Rubik’s cube which are not center facets - i.e., the ”movable” facets. Let V denote the subset of facets which belong to some corner subcube, E the subset of facets which belong to some edge subcube. Let G denote the Rubik’s cube group. As noted above, G acts on X, V , E. The action of G on X induced an equivalence relation as follows: we say that a facet f1 is ”equivalent” to a facet f2 if there is an element of G (i.e., a move of the Rubik’s cube) which sends one facet to the other. By exercise 5.8.2, there are exactly two equivalence classes, or orbits, of G in X: V and E. In particular, the action of G on V is transitive and the action of G on E is transitive.

5.9

Cosets

Let G be a group and H a subgroup of G. For g belonging to G, the subset g ∗ H of G is called a left coset of H in G and the subset H ∗ g of G is called a right coset of H in G.

5.9. COSETS

113

Exercise 5.9.1 If H is finite, show |H| = |g ∗ H| = |H ∗ g|. Exercise 5.9.2 If X is a left coset of H in G and x is an element of G, show that x ∗ X is also a left coset of H in G. Notation: The set of all left cosets is written G/H and the set of all right cosets of H in G is denoted H\G. These two sets don’t in general inherit a group structure from G but they are useful none-the-less. (G/H is a group with the ”obvious” multiplication (g1 ∗ H) ∗ (g2 ∗ H) = (g1g2 ) ∗ H if and only if H is a ”normal” subgroup of G - we will define ”normal” below.) As an example of their usefulness, we have the following relationship between the orbits and the cosets of the stabilizers. Lemma 104 Let G be a finite group acting on a set X. Then |G ∗ x| = |G/stabG (x)|, for all x belonging to X. proof: The map g ∗ stabG (x) 7−→ g ∗ x defines a function f : G/stabG (x) → G ∗ x. This function is a bijection since it is both and injection (Exercise: Check this) and a surjection (Exercise: Check this). 2 Exercise 5.9.3 Let G be the group of symmetries of the square. Using the notation above, compute G/ < g3 > and G ∗ x0 . Theorem 105 (Lagrange): If G is a finite group and H a subgroup then |G/H| = |G|/|H|. Corollary 106 If H, G are as above then the order of H divides the order of G. proof of Theorem: Let X be the set of left cosets of H in G and let G act on X by left multiplication. Apply the previous lemma with x = H. 2

114

CHAPTER 5. GROUPS, I

Exercise 5.9.4 Let G = S3 , the symmetric group on 3 letters, and let H =< s1 >, in the notation of §5.1 above. (a) Compute |G/H| using Lagrange’s Theorem. (b) Explicitly write down all the cosets of H in G. Definition 107 : Let H be a subgroup of G and let C be a left coset of H in G. We call an element g of G a coset representative of C if C = g ∗ H. A complete set of coset representatives is a subset of G, x1 , x2 , ..., xm , such that G/H = {x1 ∗ H, ..., xm ∗ H}, without repetition (i.e., all the xi ∗ H are disjoint). Exercise 5.9.5 For g1 , g2 ∈ G, define g1 ∼ g2 if and only if g1 and g2 belong to the same left coset of H in G. (a) Show that ∼ is an equivalence relation. (b) Show that the left cosets of H in G partition G.

5.10

Dimino’s Algorithm

We saw in an earlier chapter an algorithm for computing all the elements of a permuation group G. We shall discuss a more efficient algorithm for doing this in this section. For more details, see [Bu]. Notation: Let S = {g1 , g2 , ..., gn } be a set of generators for a permutation group G. Let S0 = ∅, Si = {g1 , ..., gi }, G0 = {1}, Gi =< Si >= the group generated by the elements in Si , for 1 ≤ i ≤ n. Algorithm (inductive step): Input: The generators S of G and a list L of all the elements of the permutation subgroup Gi−1 . Output: A list L of elements of Gi and a list C of coset representatives of Gi /Gi−1 .

5.10. DIMINO’S ALGORITHM

115

C = {1} for g in C do for s in S_i do if s*g not in L then C = C union {s*g} L = L union s*g*G_{i-1} endif endfor endfor Algorithm (Dimino): Input: The generators S of G Output: A list of elements of G (Initial case S_1 = ) order = 1, element[1] = 1, g = g1 while g 1 do order = order + 1 element[order] = g g = g*g1 endwhile

(General case) for i from 2 to n do endfor Example 108 Let G = S3 =< s1 , s2 >. We use Dimono’s algorithm to list all the elements of G. We have G0 = {1} < G1 =< s1 >< G2 = G. First, we list the elements of G1 =< s1 >. Since s1 = (1 2), it is order 2, so G1 = {1, s1 }.

116

CHAPTER 5. GROUPS, I

This is our list L which we will apply the ”inductive step” of Dimino’s algorithm to (with i = 2). We start with C = {1}. Now we look at the left cosets of G1 in G2 = G. We have (with g = 1, s = s1 ) s1 ∗ G1 = G1 , so we don’t increase the size of C or L. Next, we have (with g = 1, s = s2 ) s2 ∗ G1 = {s2 , s2 ∗ s1 } 6= G1 , so L = {1, s1 , s2 , s2 ∗ s1 }, C = {1, s2 }. Next, we have (with g = s2 , s = s1 ) s1 ∗ s2 ∗ G1 = {s1 ∗ s2 , s1 ∗ s2 ∗ s1 } 6= G1 . (We know s1 ∗ s2 ∗ G1 6= G1 since neither of the two elements in s1 ∗ s2 ∗ G1 is the identity.) Thus, we increase L, C: L = {1, s1 , s2 , s2 ∗ s1 , s1 ∗ s2 , s1 ∗ s2 ∗ s1 }, and C = {1, s2 , s1 ∗ s2 }. We know we may stop here since we know |S3 | = 6 but the algorithm still has one more statement to execute. Next, we have (with g = s2 , s = s2 ) s2 ∗ s2 ∗ G1 = G1 , so we don’t increase the size of C or L (as expected). This step terminates the algorithm and S3 = L. Exercise 5.10.1 Perform Dimino’s algorithm on S4 =< s1 = (1 2), s2 = (2 3), s3 = (3 4) > .

5.11

Permutations and campanology

This section is based on a capstone project of S. Robinson [Rob]. Standing outside of Westminster Abbey as the bells chime, the result you hear may actually be much more mathematical, than musical. While your ears think they detect melody,they are being deceived. The bells are not being rung in melody at all; in actuality, they are being rung in permutations ([Wh], p771). Since the seventeenth century, and possibly before, cathedral bells in England have been rung by such permutations or changes ([Wh],

5.11. PERMUTATIONS AND CAMPANOLOGY

117

p771). The art and study of such bell ringing is referred to as campanology. While campanology had been around for at least a century before the formalization of what is now known as group theory, elements of group theory are implicit in campanology. Fabian Stedman, referred to as ”one of the ’fathers of bell ringing,’” is conjectured by Arthur White, to have been, perhaps, the first group theorist ([Wh], p771). Born in 1640 to Reverend Francis Stedman, Fabian Stedman’s connections with campanology took root at the early age of 15 when he moved to London to work as an apprentice to a Master Printer. While in London Stedman joined a bell-ringing society known as the Scholars of Cheapside and served as the societies treasurer in 1662. In 1664, Stedman went on to join another bell-ringing society known then as the ”Society of Colledg(sic) Youths,” which has since been renamed the Ancient Society of College Youths and is still in existence today ([Wh], p771). Stedman remained with the society becoming Steward of College Youths in 1677 and eventually in 1682 Master of the Society. Stedman’s major contributions to campanology are reflected in his efforts on Tintinnlogia and Campanalogia, the first two books published on the subject, in 1668 and 1677, respectively ([Wh], p771). To leap right into a discussion of Stedman’s work, campanology, or group theory even in their most general terms without some cursory definitions would be futile. Below is a glossary of a few essential terms: • Transposition: a cycle (i,j) of length 2 which interchanges i and j • Change: the swapping of one or more disjoint pairs of adjacent bells • Plain change: involves swapping one pair of adjacent bells only • Cross change: involves more than one swapping pair of bells • Round: A unique ordering of the bells (i.e. (1 2 3 4...n)) The following picture was taken from [Wa], page 71.

118

CHAPTER 5. GROUPS, I

In the beginning, change ringing concerned itself with a single row of bells whose order could be denoted by (1, 2, 3, ..., n). Considering the case where n = 6 the concepts of plain and cross changes can be understood more clearly. If we use only plain changes we can generate permutations of the bells as follows: 1 2 3 4 5 6 2 1 3 4 5 6 2 1 4 3 5 6 2 1 4 3 6 5, It should be fairly obvious on inspection that the first plain change swaps 1 and 2, the second swaps 3 and 4, and the third swaps 5 and 6. Considering the same set of six bells acted upon by a cross change, the same result is achieved in one change, as seen below: 1 2 3 4 5 6 2 1 4 3 6 5. More useful and interesting patterns can be generated by combining plain

5.11. PERMUTATIONS AND CAMPANOLOGY

119

and cross changes. The plain lead on four bells is one of the most simplistic patterns and was devised sometime around 1621 by alternating consecutive cross and plain changes as seen below: 1 2 2 4 4 3 3 1 1

2 1 4 2 3 4 1 3 2

3 4 1 3 2 1 4 2 3

4 3 3 1 1 2 2 4 4

It is easy to see that the pattern which defines the plain lead on four bells is nothing more than a cross change followed by a plain change on the middle two bells until we reach the round, which is where we started. Generating the plain lead on four bells is analogous algebraically to generating the dihedral group on four elements, D4 . We begin by representing the cross change as a = (1, 2)(3, 4) which swaps the first two and last two bells and representing the plain change as b = (2, 3) which swaps the middle pair. Algebraically, D4 is generated by multiplication. We begin with the first element in the group, a. To generate the next element in the group we multiply this first element by b. To generate the third element we simply multiply this second term, ab, by a to get aba. Continuing on in this manner we multiply alternately by a then b to generate the dihedral group D4 = {a, ab, aba, (ab)2 , a(ab)2 , (ab)3 , (ab)3 a, (ab)4 }. Since (ab)4 yields the round, we say (ab)4 = 1 and D4 = {1, a, ab, aba, (ab)2 , a(ab)2 , (ab)3 , (ab)3 a}. While this simple example illustrates the implicit elements of group theory which seem to be at the heart of bell ringing, moving on to a more complex example illuminates perhaps more significant implications. We turn our attention now to the composition which is commonly referred to as Plain Bob Minimus. Plain Bob Minimus begins at the round and ends at the round (1, 2, 3, 4) and contains all possible permutations of these four bells. Calling the earlier definitions to mind, it should be evident that generating the Plain Bon Minimus composition is equivalent algebraically to generating the symmetric group on 4 elements, S4 , which is shown: 1 2 3 4

1 3 4 2

1 4 2 3

120

CHAPTER 5. GROUPS, I 2 2 4 4 3 3 1

1 4 2 3 4 1 3

4 1 3 2 1 4 2

3 3 1 1 2 2 4

3 3 2 2 4 4 1

1 2 3 4 2 1 4

2 1 4 3 1 2 3

4 4 1 1 3 3 2

4 4 3 3 2 2 1 1

1 3 4 2 3 1 2 2

3 1 2 4 1 3 4 3

2 2 1 1 4 4 3 4

We can now analyze this composition as we did D4 . We begin first by letting a = (1, 2)(3, 4) and b = (2, 3) represent possible changes between rows. If we look at the first column of the Plain Bob Minimus composition, we see that it is nothing more than the dihedral group, D4 , which is a subgroup of S4 . To generate the second column of S4 we introduce a c = (3, 4) and we simplify our notation by letting k = (ab)3 ac. Multiplying through we generate the second column, {k, ka, kab, kaba, k(ab)2 , ka(ab)2 , k(ab)3 , k(ab)3 a}. Almost immediately we should realize that this is the left coset kD4 . Employing c again to obtain the third column yields k 2 D4 , which is the final left coset since multiplication by c a third time brings us to rounds. The generation of the Plain Bob Minimus shows that S4 can be expressed as the disjoint union of cosets of the subgroup D4 , that is, also stated, the cosets of D4 in S4 partition S4 . There is an important generalization of this fact, which states: Theorem 109 For any group G and any subgroup H, the cosets of H in G partition G. Now, since we chose a = (1, 2)(3, 4), b = (2, 3), and c = (3, 4), where b and c are obviously by definition 2-cycle or transpositions and a is the product of two such 2-cycles or transpositions, we have shown a further result, that each element of S4 can be written as a product of 2-cycles. More generally, we can state the following theorem: Theorem 110 Let f be a member of Sn , i.e., let f be any permutation of degree n. Then f can be written as a product of transpositions. To sketch a proof of this theorem (following [G]), we need only to recall that:

5.11. PERMUTATIONS AND CAMPANOLOGY

121

Theorem 111 Every permutation of Sn can be written uniquely as a product of disjoint cycles. Recall that any cycle can be written as a product of transpositions as below: (a1 , a2 , ..., ak ) = (a1 , ak )(a1 , ak−1 )...(a1 , a2 ). We see that since any permutation can be written in terms of cycles and any cycle can be written as product of transposition, it follows that every permutation of Sn can be written as a product of transpositions. Considering both the plain lead on four bells and the Plain Bob Minimus composition, it is obvious that group theory is latent in the study of campanology. As White concludes in his essay, he is not suggesting ”that Fabian Stedman was using group theory explicitly, but rather that group theoretical ideas were implicit in (Stedman’s) writings and compositions” ([Wh], 778). Whether we can consider Stedman the first group theorist, then, is unclear; what is clear, however, is that when we hear the bells of Westminster Abbey chime, we are hearing mathematics, not melody.

122

CHAPTER 5. GROUPS, I

Chapter 6 Graphs and ”God’s algorithm” ”...O, cursed spite, that ever I was to set it right!” Hamlet, Act 1, scene 5 In this chapter we introduce a graphical interpretation of a permutation group, the Cayley graph. This is then interpreted in the special case of a group arising from a permutation puzzle. To begin, what’s a graph? A graph is a pair of countable sets (V, E), where • V is a countable set of singleton elements called vertices, • E is a subset of unordered pairs {{v1 , v2 } | v1 , v2 ∈ V } called edges. A graph is drawn by simply connecting points representing vertices together by a line segment if they belong to the same edge. A digraph, or directed graph, is a pair of countable sets (V, E), where • V is a countable set of vertices, • E is a subset of ordered pairs {(v1 , v2 ) | v1 , v2 ∈ V } called edges. A digraph is drawn by simply connecting points representing vertices together by an arrow if they belong to the same edge (v1 , v2 ), the araow originating at v1 and arrowhead pointing to v2 . 123

124

CHAPTER 6. GRAPHS AND ”GOD’S ALGORITHM”

If e = {v1 , v2 } belongs to E then we say that e is an ”edge from v1 to v2 ” (or from v2 to v1 ). If v and w are vertices, a path from v to w is a finite sequence of edges beginning at v and ending at w: e0 = {v, v1 }, e1 = {v1 , v2 }, ..., en = {vn , w}. If there is a path from v to w then we say v is connected to w. We say that a graph (V, E) is connected if each pair of vertices is connected. The number of edges eminating from a vertex v is called the degree (or ”valence”) of v, denoted degree(v). Example 112 : If V = {a, b, c},

E = {{a, b}, {a, c}, {b, c}},

then we may visualize (V, E) as * c / \ / \ a * --- * b Each vertex has valence 2. Definition 113 : If v and w are vertices connected to each other in a graph (V, E) then we define the distance from v to w, denoted d(v, w), by d(v, w) =

min #{edges in a path from v to w}

v,w∈V connected

By convention, if v and w are not connected then we set d(v, w) = ∞. The diameter of a graph is the largest possible distance: diam((V, E)) = max d(v, w). v,w∈V

In the above example, the diameter is 1.

6.1. CAYLEY GRAPHS

6.1

125

Cayley graphs

Let G be a permutation group, G =< g1 , g2 , ..., gn > < SX . The Cayley graph of G with respect to X = {g1 , g2 , ..., gn } is the graph (V, E) whose vertices V are the elements of G and whose edges are determined by the following condition: if x and y belong to V = G then there is an edge from x to y (or from y to x) if and only if y = gi ∗ x or x = gi ∗ y, for some i = 1, 2, ..., n. The Cayley digraph of G with respect to X = {g1 , g2 , ..., gn } is the digraph (V, E) whose vertices V are the elements of G and whose edges are determined by the following condition: if x and y belong to V = G then there is an edge from x to y if and only if y = x ∗ gi , for some i = 1, 2, ..., n. Exercise 6.1.1 Show that the Cayley graph of a permutation group is connected. Lemma 114 Let ΓG = (V, E) denote the Cayley graph associated to the permutation group G =< g1 , g2 , ..., gn >. Let N = |{g1 , g1−1 , g2 , g2−1 , ..., gn , gn−1 }|. Then, for all v ∈ V , degree(v) = N . proof: Assume not. Then there is a v ∈ V = G with either (i) degree(v) < N , or (ii) degree(v) > N . First, we note that, for each h ∈ {g1 , g1−1 , g2 , g2−1 , ..., gn , gn−1 }, the set {v, h ∗ v} is an edge of ΓG . This follows from the definition of the Cayley graph. If r = degree(v) > N then, by definition of the Cayley graph, there are distinct v1 , ..., vr ∈ V with v = hi ∗ vi , for all 1 ≤ i ≤ r, where the h1 , ..., hr are distinct elements of {g1 , g1−1 , g2 , g2−1 , ..., gn , gn−1 }. This contradicts the definition of N . If r = degree(v) < N then, by definition of the Cayley graph, there are distinct hi , hj in {g1 , g1−1 , g2 , g2−1 , ..., gn , gn−1 } such that hi ∗ v = hj ∗ v. Since V is a group and we may cancel the v’s from the two sides of the equation hi ∗ v = hj ∗ v, contradicting the assumption that hi is distinct from hj . 2

126

CHAPTER 6. GRAPHS AND ”GOD’S ALGORITHM”

Example 115 : Let G =< s1 , s2 >= S3 , where s1 = (1 2), and s2 = (2 3). Then the Cayley graph of G with respect to X = {s1 , s2 } may be visualized as

Example 116 : Let G =< s1 , s2 >= S3 , where s1 = (1 2), and s2 = (2 3). Then the Cayley digraph of G with respect to X = {s1 , s2 } may be visualized as

6.1. CAYLEY GRAPHS

127

Exercise 6.1.2 Construct the Cayley graph of C4 , the cyclic group, with respect to the generator s = (1, 2, 3, 4). Exercise 6.1.3 Construct the Cayley graph of S4 , the symmetric group on four letters, with respect to the generators s1 = (1 2), s2 = (2 3) and s3 = (3 4). Exercise 6.1.4 Construct the Cayley digraph of S3 with respect to the generators f = (1, 3), r = (1, 2, 3). (Show, in particular, that f, r do indeed generate S3 .) Example 117 : Let G =< R, L, U, D, F, B > < S54 be the group of the 3 × 3 Rubik’s cube. Each position of the cube corresponds to an element of the group G (i.e., the move you had to make to get to that position). In other words, each position of the cube corresponds to a vertex of the Cayley graph. Each vertex of this graph has valence 12

128

CHAPTER 6. GRAPHS AND ”GOD’S ALGORITHM”

Exercise 6.1.5 Check this. Moreover, a solution of the Rubik’s cube is simply a path in the graph from the vertex associated to the present position of the cube to the vertex associated to the identity element. The number of moves in the shortest possible solution is simply the distance from the vertex associated to the present position of the cube to the vertex associated to the identity element. The diameter of the Cayley graph of G is the number of moves in the best possible solution in the worst possible case.

6.2

God’s algorithm

Problem: Let G be the group of a permutation puzzle. Find the diameter of the Cayley graph of G. This problem is unsolved for must puzzles (including the 3 × 3 Rubik’s cube) and appears to be very difficult computationally. The cases where it is known include (with no attempt at completeness) the following: puzzle pyraminx 2 × 2 Rubik’s cube

diameter 11 (not including tip moves) 14

For the 2 × 2 Rubik’s cube, see [CFS]. Problem: Let G be the group of a permutation puzzle and let v be a vertex in the Cayley graph of G. Find an algorithm for determining a path from v to the vertex v0 associated to the identity having length equal to the distance from v to v0 . This problem is much harder. The algorithm, if it exists, is called God’s algorithm. A good reference for recent progress on God’s algorithm for various Rubik’s cube-like puzzles may be found on Mark Longridge’s www page [Lo]. Exercise 6.2.1 Find the Cayley graph of the ”sliced squared” group G =< MR2 , MF2 , MD2 >, where MR is the middle slice move which turns the middle slice parallel to the right face clockwise 90 degrees (with respect to the right face). Find the diameter of this graph.

6.2. GOD’S ALGORITHM

129

Let Γ be a graph. A Hamiltonian circuit on Γ is a sequence of edges forming a path in Γ which passes through each vertex exactly once. (If you think of the vertices as cities and the edges as roads then a Hamiltonian circuit is a tour visiting each city exactly once.) The following unsolved problem was first mentioned in this context (as far as I know) by A. Schwenk: Problem: Let G be the group of the 3 × 3 Rubik’s cube puzzle. Does the Cayley graph of G have a Hamiltonian circuit? In other words, can we (in principle) ”visit” each possible position of the Rubik’s cube exactly once, by making one move at a time using only the basic generators R, L, U, D, F, B? This is a special case of a more general unsolved problem: For an arbitary permutation group with more than two elements, it is not known if the Cayley graph is Hamiltonian. An example of one where it is known is the following:

Example 118 Let G be the group Sn with generators given to be the set of all transpositions: G = Sn ,

X = {(i, j) | 1 ≤ i < j ≤ n}.

(There are many more transpositions than necessary to generate Sn since the subset of transpositions of the form (i, i + 1), 1 ≤ i ≤ n − 1, suffice to generate Sn [R].) The algorithm of Steinhaus (see §3.3) shows that there is a Hamiltonian circuit in the Cayley graph of Sn with respect to X.

The reader interested in more examples is referred to [CG].

6.2.1

History

Sir William Hamilton, an Irish prodigy of the 18th century, originated the problem of finding Hamiltonian paths by patenting a game called the Icosian game or the Hamilton game. The idea is to find a Hamiltonian path around the vertices of the icosahedron. A picture of the orginal game is:

130

6.3

CHAPTER 6. GRAPHS AND ”GOD’S ALGORITHM”

The graph of the 15 puzzle

This section, which is based on [Mc], discusses the 15 puzzle from the graphtheoretical point of view following [W]. The 15 puzzle was introduced in §4.1 above. The object of the puzzle was to order the pieces from one to fifteen from left to right, top to bottom, as shown in the solved position:

6.3. THE GRAPH OF THE 15 PUZZLE

131

To solve a mixed up puzzle, one would slide the squares around in the puzzle. In order to do this you must slide a numbered square into the place of the space. We could represent this mathematically by saying that this is a transposition of that numbered square and the blank. If we label each space in the puzzle in the above Figure, as a vertex, and label the vertices numerically, then the resulting graph is represented by

132

CHAPTER 6. GRAPHS AND ”GOD’S ALGORITHM”

We will let 16 denote the blank and call this graph Γ. The only legal moves of the puzzle are transpositions of the 16th vertex and a vertex that is adjacent to it. Therefore, any permutation of the vertices produces a labeling on Γ.

6.3.1

General definitions

Now let Γ be a simple graph with the vertex set V (Γ) of cardinality N . (In the above example N = 16.) By a labeling we mean the placement of the numbers one through N on distinct vertices of Γ, where N denotes the blank. In other words, a labeling on Γ is a bijective mapping f : V (Γ) → {1, 2, ..., N }. Two labelings f, g on Γ are adjacent if and only if g is a result of a single transposition on f of vertex N with a vertex adjacent to N on f . In other words, f and g are adjacent if they differ by one legal move of the puzzle. From Γ, we make a new graph puz(Γ) as follows [W]: the vertex set V (puz(Γ)) contains all labelings on Γ, and two vertices in puz(Γ) are joined by an edge if the associated labelings are adjacent. For example, the labelings below are adjacent:

6.3. THE GRAPH OF THE 15 PUZZLE

133

We can consider a sequence of moves on Γ to be a path p, such that p = (x0 , x1 , x2 , ..., xn ) where the x0i s are vertices of Γ, and xi and xi−1 are adjacent. Such a path p is said to be from x0 (its initial vertex) to xn (its terminal vertex). The path p is simple if x0 , x1 , x2 , ..., xn are distinct. If x0 = xn then (a not necessarily simple path) p is called a closed path based at x0 . Let x0 be a fixed vertex of Γ. The set of paths based at x0 forms a group (under composition of paths) called the homotopy group of Γ based at x0 , denoted Γ(x0 ). Now suppose we paint the blocks of the 14-15 puzzle in a checkerboard pattern:

134

CHAPTER 6. GRAPHS AND ”GOD’S ALGORITHM”

In this arrangement the blank would start on a white square. If we were to move the blank up, then it is now on a black square. That is one transposition, therefore the movement is odd. If we then move the blank to the left, the blank would be on a white square. This is a total of two transpositions; therefore, the movement is even. After three transpositions the blank would be on a black square, therefore it would be an odd permutation. Therefore if the blank ends on a white square, an even permutation has occurred. If the blank ends on a black square, an odd permutation has occurred. A legal position of the 15-Puzzle is any sequence of legal transpositions starting from the solved position such that the blank ends up in the bottom right-hand corner. Each such position corresponds to a permutation of the 15 numbered vertices and hence to an element of the symmetric group S15 . The set of all such permutations (arising as a sequence of transpositions) in S15 forms a group called the group of the 15-Puzzle. Note that the group of the 15-puzzle is isomorphic to the homotopy group of the 15 puzzle graph based at the ”blank” vertex. If we assign the number 16 to the blank, then we can see that we can arrange the pieces of the puzzle in 16! different ways. However, if we take only legal positions of the 15-Puzzle, then we are fixing one of the pieces. As a result the number of ways to permute the rest of the pieces, with the blank on the white square at the bottom right-hand corner, is at most 15!.

6.3. THE GRAPH OF THE 15 PUZZLE

135

All such permutations have to be even, by the checkerboard analysis above. 15!/2 is the number of even permutations of 15 elements (there are an equal number of even and odd permutations). From this we see that the 15-puzzle has 15!/2 possible legal positions.

Theorem 119 The positions with the empty space at the bottom right that can be reached from the start position of the 15-Puzzle by shifting tiles are in a bijective correspondence with the 15!/2 = 1, 307, 674, 368, 000 even permutations of the numbers from 1 to 15.

Remark 9 The 14-15 Puzzle cannot be solved, because it is an odd permutation. It only has one transposition, 14 interchanged with 15.

proof: If we label the empty space 16, then every possible position of the puzzle may be regarded as an element of the symmetric group S16 and an element of puz(Γ). There are 16! elements in S16 , and 16! vertices of puz(Γ). With the argument earlier we can show that all legal positions of the puzzle are obtained by an even number of transpositions. Therefore, all legal moves are even permutations of the puzzle. Now we must show that there is a certain 3-cycle in the group of the 15-Puzzle. For example, if we shift the three pieces surrounding the empty space around in a circle following the order of moves south-east-north-west then the three-cycle (11, 12, 15) is produced, as indicated in the previous Figures combined with

136

CHAPTER 6. GRAPHS AND ”GOD’S ALGORITHM”

It can be shown that if you fix the 11 and 12 pieces, then any other piece can take the place of the 15 by following one of the cycles below:

6.3. THE GRAPH OF THE 15 PUZZLE

137

By Lemma 159, such 3-cycles generate A15 . This proves the theorem. Alternatively, with some work we can show that any number can replace the 11 in the three-cycle, and we can show that any other number can replace the 12. From this we can conclude that any three-cycle can be formed. Since every even permutation is a combination of three-cycles (by Proposition 158), every even permutation of the 15-Puzzle can be reached. 2 With this information, we can make a generalization to rectangular puzzles of size m × n with m > 1 and n > 1: Theorem 120 The group of an m × n rectangular puzzle is the alternating group Amn−1 . The proof of this is similar to the proof for the 4 × 4 puzzle, if m > 3 and n > 3. (The special cases when 1 < m < 4 or 1 < n < 4 must be treated separately). The size of the alternating group is given by (mn − 1)!/2.

138

6.4

CHAPTER 6. GRAPHS AND ”GOD’S ALGORITHM”

Remarks on applications, NP-completeness

Cayley graphs have been used by computer scientists to model interconnection networks for parallel processors (see [CFS], [CG] for some references). The problem of finding an efficient algorithm for the shortest solution to the m × n puzzle is difficult. It amounts to finding the shortest path between two points in a graph which is, in general, a difficult problem computationally [GJ]. There is a class of problems called ”NP-complete” problems. Without getting into precise details which would take us too far afield, this is a class of problems which are in some sense ”equally hard” to solve. If you can find a ”polynomial time” algorithm to solve one then you can find one to solve any other problem in that class as well. For example, [GJ] and [BCG] have a list of games and puzzles whose solutions are NP-complete problems.

Chapter 7 Symmetry groups of the Platonic solids ”Plato said God geometrizes continually.” Plutarch Convivialium disputationum, liber 8,2 ”We do not listen with the best regard to the verses of a man who is only a poet, nor to his problems if he is only an algebraist; but if a man is at once acquainted with the geometric foundation of things and with their festal splendor, his poetry is exact and his arithmetic musical.” Ralph Waldo Emerson Society and Solitude chapter 7, Work and Days This chapter requires a little more mathematical sophistication from the reader than the earlier chapters. However, the exercises are (I think) choosen to be doable.

7.1

Descriptions

The ”Platonic solids” are the 5 regular polyhedrons: 139

140 CHAPTER 7. SYMMETRY GROUPS OF THE PLATONIC SOLIDS polyhedron tetrahedron hexahedron octahedron dodecahedron icosahedron Here:

# faces 4 6 8 12 20

# vertices 4 8 6 20 12

# edges 6 12 12 30 30

group T O O I I

p,q 3,3 4,3 3,4 5,3 3,5

p, called the face degree, denotes the number of edges bounding each face, q, called the vertex degree, denotes the number of faces meeting each vertex. A vertex of one of these solids is therefore specified by the q-tuple of faces meeting that vertex. We saw several examples of this already when we specified notation for the movements of the associated Rubik’s cube-like puzzles in chapter 5. These solids may be drawn in rectangular coordinates using polyhedron coordinates tetrahedron (1,1,1), (1,-1,-1), (-1,-1,1), (-1,1,-1) hexahedron (1,1,1), (1,1,-1), (1,-1,1), (-1,1,1), (1,-1,-1), (-1,1,-1), (-1,-1,1), (-1,-1,-1) octahedron (1,0,0), (0,0,1), (0,1,0), (-1,0,0), (0,-1,0), (0,0,-1) dodecahedron (0,±φ−1 ,±φ), (±φ−1 ,±φ,0), (±φ,0,±φ−1 ), (±1,±1,±1), icosahedron (1,0,φ), (1,0,−φ), (-1,0,φ), (-1,0,−φ), (0,φ,1), (0,φ,-1), (0,−φ,1), (0,−φ,-1), (φ,1,0), (φ,-1,0), (−φ,1,0), (−φ,-1,0) where φ denotes the golden ratio. If P1 , P2 , P3 are three vertices of an icosahedron which form a triangular face then (P1 + P2 + P3 )/3 forms a vertex of the dual dodecahedron and every vertex of the dual dodecahedron arises in this way. The three ”Platonic groups” (the group of ”symmetries” of these figures) will be described below. Their names: T = symmetric group of the tetrahedron = tetrahedral group, O = symmetric group of the octahedron (or cube) = octahedral group, I = symmetric group of the icosahedron (or dodecahedron) = icosahedral group.

7.2. BACKGROUND ON SYMMETRIES IN 3-SPACE

7.2

141

Background on symmetries in 3-space

This subsection presents, with some proofs, background on isometries in 3 dimensions necessary for understanding the symmetry groups of the Platonic solids. We fix once and for all the ”right-hand-rule” orientation in 3-space. We call a distance-preserving transformation in 3-space which fixes the origin a symmetry of 3-space. We say that such a symmetry is orientation preserving if it preserves the right-hand rule orientation. Example 121 : Let s : R3 → R3 denote the function which takes each vector v belonging to R3 and returns its reflection s(v) about the yz-plane. This is not orientation preserving since it reverses the direction of a counterclockwise moving circular path in the yz-plane. In terms of rectangular coordinates, s(x, y, z) = (−x, y, z). Let R3 = {(x, y, z) | x, y, z real numbers} denote 3-space. We also write this, when convenient, as column vectors 



x  3 R = { y   | x, y, z real} z The distance function on R3 is the function q

d(~v1 , ~v2 ) =

(x1 − x2 )2 + (y1 − y2 )2 + (z1 − z2 )2

where ~v1 = (x1 , y1 , z1 ), ~v2 = (x2 , y2 , z2 ). This may be expressed in terms of q the inner product ~v1 ·~v2 = x1 x2 +y1 y2 +z1 z2 as d(~v1 , ~v2 ) = (~v1 − ~v2 ) · (~v1 − ~v2 ). Conversely, the polarization identity: 1 ~v1 · ~v2 = (||~v1 + ~v2 ||2 + ||~v1 − ~v2 ||2 ) 2 allows one to recover the value of the inner product from the knowledge of the values of the distance function. We call a function f : R3 → R3 an isometry if it satisfies d(f (v1 ), f (v2 )) = d(v1 , v2 )

142 CHAPTER 7. SYMMETRY GROUPS OF THE PLATONIC SOLIDS for all v1 and v2 belonging to R3 . We want to understand isometries a little better since they will preserve distances (and, in particular, preserve the shapes of solids) and therefore provide us with the kinds of symmetries of 3-space we want to consider. We can construct isometries using certain types of 3 × 3 matrices. (Appendix 1 of this chapter gives a little background on matrices.) Lemma 122 If A is a 3 × 3 matrix then the function A : R3 → R3 is an isometry if and only if At ∗ A = I3 , where At denotes the transpose identity matrix (obatined by flipping the entries of A about the diagonal. Remark 10 In particular, if A is an isometry then det(A)2 = det(At ) det(A) = det(At ∗ A) = det(I3 ) = 1. proof: The distance function is preserved if and only if the dot product function is preserved. (This is a consequence of the ”polarization identity” - see the appendix.) Let m(~v , w) ~ = ~v · w, ~ where · denotes the vector dot t product. Since m(A~v , B w) ~ = ~v · (A ∗ B)w, ~ we have m(A~v , Aw) ~ = m(~v , w), ~

∀v, w ∈ R3

if and only if ~v · (At ∗ B)w ~ = ~v · w, ~

∀v, w ∈ R3

if an only if At ∗ A = I3 . 2 You may have been wondering how one could construct an isometry. This lemma gives us lots of examples. Example 123 A rotation matrix in 3-dimensions may be written in the form 



r11 r12 sin(θ) sin(ψ)  r21 r22 sin(θ) cos(ψ)  R(φ, θ, ψ) =   sin(φ) sin(θ) − sin(θ) cos(φ) cos(θ), where

r11 = cos(φ) cos(ψ) − cos(θ) sin(φ) sin(ψ), r12 = sin(φ) cos(ψ) + cos(θ) cos(φ) sin(ψ), r21 = − cos(φ) sin(ψ) − cos(θ) sin(φ) cos(ψ), r22 = − sin(φ) sin(ψ) + cos(θ) cos(φ) cos(ψ).

and where the angles φ, θ, ψ are the ”Euler angles”. This represents the rotation of 3-space obtained by the following sequence of rotations: rotate by

7.2. BACKGROUND ON SYMMETRIES IN 3-SPACE

143

angle ψ about the z-axis, rotate by the angle θ about the x-axis (0 ≤ θ ≤ π), then rotate by angle φ about the z-axis again. Although this is an interesting fact due to its explicitness, we shall not use this expression. Question: Are there any isometries which do not come from matrices as in the above lemma? Yes: any translation gives rise to an isometry. Question: Are there any examples of isometries which do not arise from a composition of a translation and an orthogonal matrix? No: the following theorem classifies all the isometries. Theorem 124 A function f : R3 → R3 is an isometry fixing the origin if and only if f is left multiplication by an orthogonal matrix. This will not be proven here (see Artin [Ar], chapter 4, section 5, Proposition 5.16). As a consequence of this lemma, we see that if the matrix A gives rise to as isometry then det(A) is either equal to 1 or -1 (since det(A)2 = det(At ∗ A) = det(I3 ) = 1). In particular, the determinant of such a matrix is non-zero, so the matrix is invertible. Lemma 125 The set of all 3×3 matrices A such that the function A : R3 → R3 is an isometry forms a group under matrix multiplication. Exercise 7.2.1 Verify the group axioms needed to prove this lemma. Notation: This group will be denoted O3 (R) and called the orthogonal group of R3 . We denote by SO3 (R) the following subset SO3 (R) = {A ∈ O3 (R) | det(A) = 1}. which is called the special orthogonal group of R3 . Lemma 126 SO3 (R) is a subgroup of O3 (R). Exercise 7.2.2 Verify the group axioms for SO3 (R). It is known that the number of cosets in O3 (R)/SO3 (R) is 2. In fact, it is known that O3 (R) = SO3 (R) ∪ s ∗ SO3 (R) (disjoint union)

(7.1)

where s is the reflection in the above example (this follows from [Ar], chapter 4, section 5).

144 CHAPTER 7. SYMMETRY GROUPS OF THE PLATONIC SOLIDS Lemma 127 The isometry A in O3 (R) is orientation preserving if and only if det(A) = 1. We will not prove this lemma here.

7.3

Symmetries of the tetrahedron

Fix a tetrahedron centered at the origin, with one vertex along the z-axis. Each edge has an ”opposite” edge on the tetrahedron (which is actually perpendicular to it if you look at it straight on). Each vertex has an ”opposite” face.

There are orientation preserving symmetries (called rotations) of the tetrahedron and orientation reversing symmetries of the tetrahedron. The orientation preserving symmetries of the tetrahedron will be denoted ST . They are obtained as follows:

7.4. SYMMETRIES OF THE CUBE

145

• the 4 axes of symmetry through the centers of the faces yield 2 elements each (120 degree clockwise rotation when viewed from outside and a 240 degree rotation), for a total of 8 elements, (This ”tetrahedral symmetry” allows for the mechanical construction of the pyraminx.)

• the 3 pairs of edges (formed by an edge and its opposite) yield one element each (a 180 degree rotation), for a total of 3 elements.

These, plus the identity, give 12 elements in ST . Using the coset decomposition (7.1), we have T = ST ∪ s ∗ ST (disjoint), so |T | = 24.

Remark 11 It turns out that ST is essentially the alternating group A4 of even permutations of S4 and T is essentially S4 itself. We shall state the precise result in the next chapter.

7.4

Symmetries of the cube

We fix a cube centered about the origin in 3-space. The set of centers of the faces of a cube forms a set of vertices of an octahedron drawn inside the cube. This octahedron is called the ”dual” polyhedron.

146 CHAPTER 7. SYMMETRY GROUPS OF THE PLATONIC SOLIDS

These two polyhedra have the same symmetry group, which we denote by O. There are orientation preserving symmetries, or rotations, of the cube and orientation reversing symmetries of the cube. The orientation preserving symmetries of the cube will be denotes SO. They are obtained as follows: • the 3 axes of symmetry through the centers of the faces yield 3 elements each (90 degree clockwise rotation when viewed from outside, a 180 degree rotation, and a 270 degree rotation), for a total of 9 elements, (This ”hexahedral symmetry” allows for the mechanical construction of the Rubik’s cube.) • the 4 axes through the opposing vertices yield 2 elements each (all of order 3), for a total of 8 elements, (This ”tetrahedral symmetry” allows for the construction of the skewb [H].) • the 6 axes through the opposing mid-edge points yield 1 element each (of order 2), for a total of 6 elements.

7.4. SYMMETRIES OF THE CUBE

147

These elements, plus the identity, yield 24 elements.

Lemma 128 There are 24 orientation preserving elements in O, i.e., |SO| = 24.

The above sketch is one way to see why this is true. Here’s another proof: Let V be the set of vertices of the cube. The group SO acts on the set V . Fix a v belonging to V and let H = stabSO (v). One can check that |H| = 3 (since the only symmetry which fixes v is a rotation g about the line through v and its opposite vertex. Since g is order 3, H =< g > is order 3 as well). We have |V | = 8, so by a lemma in the previous chapter on orbits and stabilizers, we have |SO/H| = |V |. By Lagrange’s theorem, |SO| = |SO/H||H| = 8 · 3 = 24. 2 Now we know SO, what is O? Note that s, the reflection in the example in the previous section, belongs to O. Using the coset decomposition of the previous section, we have the coset decomposition O = SO ∪ s ∗ SO

(disjoint union).

We know that |s ∗ SO| = |SO| = 24, so

Lemma 129 The order of the octahedral group is |O| = 48.

Remark 12 It turns out that SO is essentially the symmetric group S4 and O is ”isomorphic to the direct product” S4 × C2 . We shall state the precise definitions and result in the next chapter.

148 CHAPTER 7. SYMMETRY GROUPS OF THE PLATONIC SOLIDS

7.5

Symmetries of the dodecahedron

The set of centers of the faces of a dodecahedron forms a set of vertices of an icosahedron drawn inside. This icosahedron is called the ”dual” polyhedron. We fix a dodecahedron in 3-space so that the vertices of the dual icosahedron are as listed in section 1 above.

7.5. SYMMETRIES OF THE DODECAHEDRON

149

Let SI denote the group of orientation preserving symmetries of the dodecahedron. Note SI is a finite subgroup of SO3 (R). Let I denote the group of all symmetries of the dodecahedron. Note I is a finite subgroup of O3 (R) and that SI is a subgroup of I. Let F denote the set of faces of the dodecahedron, so |F | = 12. SI acts on F . Lemma 130 SI acts on F transitively. We won’t prove this. If you look at a dodecahedron it follows ”by inspection”. The reason why this is useful is that it tells us that if x is any face then any other face can be obtained from x by applying some element of SI. In other words, the orbit of x is all of F : SI ∗ x = F . If x is any face then the only orientation preserving symmetries which don’t send x to a different face is a rotation by an integer multiple of 72 degrees about the line passing through the center of x and the center of its opposite face. There are, for each face x, exactly 5 distinct rotations of this type. Therefore, |stabSI (x)| = 5.

150 CHAPTER 7. SYMMETRY GROUPS OF THE PLATONIC SOLIDS By a lemma in the section on orbits, we have SI/stabSI (x) = SI ∗ x, so |SI| = |stabSI (x)||SI ∗ x| = 5 · 12 = 60. The elements of SI include: • rotation by 2 ∗ π ∗ k/5, k = 0, 1, 2, 3, 4, about the line passing through the center of a face and its opposite, (This ”dodecahedral symmetry” allows for the construction of the megaminx.) • rotation by 2 ∗ π ∗ k/3, k = 0, 1, 2, about the line passing through a vertex and its opposite, • rotation by π about the center of an edge. Subgroups include: • stabilizer of a vertex. These are all cyclic of order 3, and they are all conjugate. There are 10 distinct such subgroups since a vertex and its opposite share the same stabilizer. • stabilizer of a face. These are all cyclic of order 5, and they are all conjugate. There are 6 distinct such subgroups since a face and its opposite share the same stabilizer. • stabilizer of an edge. These are all cyclic of order 2, and they are all conjugate. There are 15 distinct such subgroups. Exercise 7.5.1 Verify all these. Remark 13 It turns out that SI is essentially the alternating group A5 of even permutations of S5 and I is ”isomorphic to the direct product” A5 × S5 . We shall state the precise result in the next chapter. For an excellent discussion of the symmetries of the icosahedron, see [Ba].

7.6. APPENDIX: SYMMETRIES OF THE ICOSAHEDRON AND S6 151

7.6

Appendix: Symmetries of the icosahedron and S6

A duad is a pair of diagonals (a diagonal is a segment from a vertex to its antipodal opposite vertex) of the icosahedron. The top of the icosahedron has 6 vertices and each diagonal must have exactly one of these 6 vertices as an endpoint. There are 12 vertices, hence 6 diagonals, hence Ã

6 2

!

= 15

different duads. Each duad determines a ”golden rectangle” (i.e.,√a rectangle whose ratio length/width is either the golden ratio φ = (1 + 5)/2 or its inverse. We may identify a duad with a pair of distinct integers {(i, j) | 1 ≤ i < j ≤ 6}, i.e., with a 2-cycle in S6 . Each element of the rotation group of the icosahedron (i.e., the group of orientation-preserving symmetries of the icosahedron) must send a duad to a duad. Each duad has 4-fold symmetry, i.e., can be send to itself in 4 ways. There are 15 duads, so there are 4 · 15 = 60 ways to send a duad to another. This is precisely the number of orientation-preserving symmetries of the icosahedron. The 18-th century mathematician Sylvester (who, though from Great Britain, once taught in Maryland) called a partitioning X = X1 ∪ ... ∪ Xn

(disjoint),

of a set X a syntheme if each of the sets Xi , 1 ≤ i ≤ n, has the same number of elements. If we take X = {set of diagonals of the icosahedron} then a syntheme is a set of three duads, no two having a diagonal in common. There are à !à ! 6 4 /3! = 15 2 2 different synthemes (the 3! since there are 3! ways to permute the duads amongst themselves). A syntheme may be represented by a coloring of the vertices on the top of the icosahedron using three colors, each for exactly two

152 CHAPTER 7. SYMMETRY GROUPS OF THE PLATONIC SOLIDS vertices. We may identify a syntheme with a product of 3 distinct 2-cycles in S6 . We partition the set of 15 duads into 5 groups of 3 as follows. (Recall each syntheme is a triple of duads.) First, pick a syntheme, A1 . Pick another syntheme A2 , so that A1 , A2 have no duad in common. Continue on in this way until you pick five synthemes A1 , .., A5 , no two of which have a duad in common. Such a choice of 5 synthemes is called a pentad. There a 6 pentads, which we label P1 , ..., P6 in any way you like. (A list of the six pentads is given in [R], chapter 7.) Any permutation of the 6 diagonals of the icosahedron gives rise to a permutation of the set of 6 pentads. Hence any permutation of the 6 diagonals of the icosahedron, which may be regarded as an element of S6 , gives rise to a permutation of the set of 6 pentads, which may also be regarded as an element of S6 . This gives a map f : S6 → S6 . This example will be discussed further in the next chapter. See also [R] and [Ba].

Chapter 8 Groups, II ”The art of doing mathematics consists in finding that special case which contains all the germs of generality.” David Hilbert Groups are analogous to molecules. We (mathematicians) want to know what they ”look like”, we want to know how to describe them, how to compare them, how to ”make” more of them, if they fall into ”families” with similar properties... Given two groups G1 , G2 , a natural question is to ask how ”similar” are they? (Exactly what is meant by ”similar” will be explained later.) We shall, in this chapter, introduce notions and techniques useful for comparing two groups. In a later chapter, we will focus on the 3 × 3 Rubik’s cube group by comparing it to ”better understood” groups.

8.1

Homomorphisms

A homomorphism between two groups is, roughly speaking, a function between them which preserves the (respective) group operations. Definition 131 Let G1 , G2 be groups, with ∗1 denoting the group operation for G1 and ∗2 the group operation for G2 . A function f : G1 → G2 is a homomorphism if and only if, for all a, b ∈ G1 , we have f (a ∗1 b) = f (a) ∗2 f (b). 153

154

CHAPTER 8. GROUPS, II

Exercise 8.1.1 Prove the following: If f : G1 → G2 is a homomorphism of groups then f (G1 ) = {g ∈ G2 | g = f (x), for some x ∈ G1 } is a subgroup of G2 . The subgroup f (G1 ) < G2 is called the image of f and is sometimes denoted im(f ). Example 132 Let G be a group and h a fixed element of G. Define f : G → G by f (g) = h−1 ∗ g ∗ h, g ∈ G. Then the following simple trick f (a ∗ b) = h−1 ∗ (a ∗ b) ∗ h = h−1 ∗ a ∗ h ∗ h−1 ∗ b ∗ h = f (a) ∗ f (b) shows that f is a homomorphism. In this case, im(f ) = G, i.e., f is surjective. Exercise 8.1.2 Let 





0 1 0  A= 1 0 0  , 0 0 1



1 0 0  B= 0 0 1   0 1 0

Now, let G =< A, B > denote the group of all matrices which can be written as any arbitrary product of these two matrices (in any order and with as many terms as you want). We have  

G = {I3 =  

0 1

  0 0

1 0

1 0 0 0 1 0



0 0 1 0  , 0 1  0   ,  1 0

  

0 0 1

1 0 0 1 0 0



0 0 0 1  , 1 0  0   ,  0 1

  

0 1 0

0 1 0 1 0 0



1 0 0 0  , 0 1   }

(You may want to try to check this as an exercise by regarding each such matrix as a permutation matrix.) Define f : G → S3 by

8.1. HOMOMORPHISMS

155

g f (g) I3 1 A s1 B s2 A∗B s1 ∗ s2 B∗A s2 ∗ s1 A ∗ B ∗ A s1 ∗ s2 ∗ s1 Show that this is a homomorphism. Example 133 The function sgn : Sn → {±1}, which assigns to each permutation its sign, is a homomorphism. One reason why this is true is because the sign of a permutation g is the determinant of the associated permutation matrix P (g). Since the determinant of the product is the product of the determinants (this is a basic result in linear algebra [JN]), we have sgn(gh) = det P (gh) = det P (g) det P (h) = sgn(g)sgn(h), for all g, h ∈ Sn . From this it follows that sgn is a homomorphism. Lemma 134 If f : G1 → G2 is a homomorphism then (a) f (e1 ) = e2 , where e1 denotes the identity element of G1 and e2 denotes the identity element of G2 , (b) f (x−1 ) = f (x)−1 , for all x belonging to G1 , (c) f (y −1 ∗1 x ∗1 y) = f (y)−1 ∗2 f (x) ∗2 f (y), for all x, y belonging to G1 , (∗1 denoting the group operation for G1 and ∗2 the group operation for G2 ). proof: (a) We have f (x) = f (x ∗1 e1 ) = f (x) ∗2 f (e1 ), for any x ∈ G1 . Multiply both sides of this equation on the left by f (x)−1 . (b) We have, by part (a), e2 = f (e1 ) = f (x ∗1 x−1 ) = f (x) ∗2 f (x−1 ). Multiply both sides of this equation on the left by f (x)−1 . Exercise 8.1.3 Prove part (c). 2 Definition 135 Let G1 , G2 be finite groups. We say that G1 embeds (or injects) into G2 if there exists an injective homomorphism f : G1 → G2 . A homomorphism f : G1 → G2 is a isomorphism if it is a bijection (as a function between sets). In this case, we call G1 and G2 isomorphic and write G1 ∼ = G2 . An isomorphism from a group G to itself is called an automorphism.

156

CHAPTER 8. GROUPS, II

The notion of an isomorphism is the notion we will use when we want to same two groups are ”essentially the same group”, i.e., one is basically a carbon copy of the other with the elements relabeled.

8.2

Homomorphisms arising from group actions

Lemma 136 Let G be a group and X a finite set. If G acts on X (on the left, resp. on the right) then there is a homomorphism G → SX given by g 7−→ φg . Conversely, if φ : G → SX is a homomorphism then φ(g) : X → X defines a (left, resp. right) action of G on X. Exercise 8.2.1 Prove this. Example 137 Let G be the Rubik’s cube group generated by the basic moves R, L, U, D, F, B. For each move g ∈ G, let ρ(g) be the corresponding permutation of the set of vertices V of the cube and let σ(g) be the corresponding permutation of the set of edges E of the cube. Let Sn denote the symmetric group on n letters and identify SV with S8 , SE with S12 . Then (a) ρ : G → S8 is a homomorphism, (b) σ : G → S12 is a homomorphism. Exercise 8.2.2 Prove this. Definition 138 Let G act on a set X. We call the action k-tuply transitive if for each pair of ordered k-tuples (x1 , x2 , .., xk ), (y1 , y2 , .., yk ) of elements belonging to X there is a g ∈ G such that yi = φg (xi ) for each 1 ≤ i ≤ k. Exercise 8.2.3 Is the Rubik’s cube group 2-transitive on the set of edge facets? The following result is one illustration of how unique the symmetric group and alternating group are. (Recall that the alternating group An was defined in example 81 above.) Theorem 139 If k > 5 and G is a group acting k-transitively on a finite set X then G is isomorphic to Sm or to An , for some m ≥ k or some n ≥ k + 2. Conversely, Sn acts n-transitively on {1, 2, ..., n} and An acts (n − 2)transitively on {1, 2, ..., n}.

8.3. EXAMPLES OF ISOMORPHISMS

157

This is proven in [R]. Remark 14 We shall discuss an example of this in §§13.6, 13.7. We shall also, in conjunction with our group-theoretical determination of the Rubik’s cube group proven later, be able to deduce from Theorem 139 the following Corollary 140 (a) The Rubik’s cube group G acts 6-transitively on the corners, leaving the edges alone. It acts 8-transitively on the corners but may permute two edges. (b) The Rubik’s cube group G acts 10-transitively on the edges, leaving the corners alone. It acts 12-transitively on the edges but may permute two corners.

8.3

Examples of isomorphisms

Example 141 Let G be the group in Exercise 8.1.2 and f : G → S3 the homomorphism. This is in fact an isomorphism. Example 142 Let H be the subgroup of the Rubik’s cube group generated by the basic move R: H =< R >. Then H ∼ = C4 (where C4 denotes the cyclic group of order 4). Example 143 Recall that to each permutation g of the set {1, 2, ..., n} we can associate a n × n permutation matrix P (g) in such a way that   

P (g)   

x1 x2 .. . xn





    =    

xg(1) xg(2) .. .

   .  

xg(n)

Here the image of i under the permutation g is denoted g(i), though in fact one plugs i into g from the left.) We let Pn denote the set of all n × n permutation matrices. This is a group under matrix multiplication. The function P : Sn → Pn , g 7−→ P (g) is an isomorphism. The proof that this is a bijection and a homomorphism was given earlier in the chapter on permutations.

158

CHAPTER 8. GROUPS, II

Example 144 From [NST], we have the following table of isomorphisms: name symmetry group of tetrahedron rotation group of tetrahedron symmetry group of octahedron rotation group of octahedron symmetry group of icosahedron rotation group of icosahedron symmetry group of regular n-gon

notation T ST O SO I SI

rotation group of regular n-gon

isomorphic to S4 A4 S4 × C2 S4 A5 × C2 A5 D2n , n odd Dn × C2 , n even Dn

Example 145 This example may be found in [B]. Let Q denote the quaternion group: Q = {1, −1, i, −i, j, −j, k, −k}, where i2 = j 2 = k 2 = −1, ij = k, jk = i, ki = j, and in general, xy = −yx for x, y belonging to i, j, k. Then Q is isomorphic to the group Q∗ =< a, b > < G, where a = F 2 ∗ MR ∗ U −1 ∗ MR−1 ∗ U −1 ∗ MR ∗ U ∗ MR−1 ∗ U ∗ F 2 , b = F ∗ U 2 ∗ F −1 ∗ U −1 ∗ L−1 ∗ B −1 ∗ U 2 ∗ B ∗ U ∗ L, via the map f : Q∗ → Q defined by φ(a) = i, φ(b) = k. The proof of this claim will be formulated in a later chapter as an exercise. (The easiest way to prove this uses ideas we haven’t yet introduced.) Exercise 8.3.1 As above, let G be a group and h a fixed element of G. Define ch : G → G by ch (g) = h ∗ g ∗ h−1 ,

g ∈ G.

Show that this is an automorphism. Definition 146 An automorphism as in the above exercise is called inner. An automorphism of G which is not of this form, for some h ∈ G, is called outer.

8.3. EXAMPLES OF ISOMORPHISMS

159

(solution to Exercise 8.3.1: To verify this, we must show that f = ch is an injective and surjective homomorphism (we drop the subscript for simplicity of notation). First, we show that f is injective. Suppose f (g1 ) = f (g2 ). Then f (g1 ∗ −1 g2 ) = 1, so that h ∗ g1 ∗ g2−1 ∗ h−1 = 1. Multiply both sides of this equation on the left by h and on the right by h−1 . We obtain g1 ∗g2−1 = 1. This implies g1 = g2 , so f is injective. Now we show f is surjective. Let g be an arbitrary but fixed element of G. Let y = h−1 ∗ x ∗ h. Then f (y) = f (h−1 ∗ x ∗ h) = h ∗ h−1 ∗ x ∗ h ∗ h−1 = x. Therefore, f is surjective. We verified previously that f is a homomorphism.) Notation: The set of all automorphisms of a group G is denoted Aut(G). The subset of inner automorphisms is denoted Inn(G) := {f ∈ Aut(G) | f = ch , some h ∈ G}, in the notation of the above 8.3.1. Exercise 8.3.2 (a) Show Aut(G) is a group with composition as the group operation. (b) Show that Inn(G) is a subgroup of Aut(G). (c) Show that the function φ : G → Inn(G) defined by φ(h) = ch is a homomorphism.

8.3.1

Conjugation in Sn

The following result will be of importance to us in a later chapter: Lemma 147 Suppose f : Sn → Sn is an inner automorphism. If g ∈ Sn is a disjoint product of cycles of length k1 , ..., kr then f (g) is a disjoint product of cycles of length k1 , ..., kr . In other words, an inner automorphism (i.e., conjugation) must ”preserve the cycle structure”. Lemma 50 , whose proof was promised earlier, follows from this. proof: Since f is inner, let it be conjugation by h ∈ Sn say, so f (g) = −1 h gh, for all g ∈ Sn . Let (i)g ∈ {1, ..., n} denote the image of i ∈ {1, ..., n}

160

CHAPTER 8. GROUPS, II

under g ∈ Sn . The lemma is a consequence of the following simple calculation: if (i)g = j then, for all 1 ≤ i ≤ n, we have ((i)h)(h−1 gh) = (j)h.

(8.1)

In other words, if g sends i 7−→ j then h−1 gh sends (i)h 7−→ (j)h. It follows that g and h−1 gh have the cycle structure. 2 Theorem 148 Two elements g, g 0 ∈ Sn are conjugate if and only if they have the same cycle structure. proof: The Lemma proves the ”only if” direction of this equivalence. Suppose that g, g 0 ∈ Sn have the same cycle structure. Write their disjoint cycle decompositions using the lexicographical ordering imposed on the lengths of the cycles occurring in the decomposition: say g = (i1 ... in1 )(in1 +1 ... in2 )...(ink +1 ... in ), g 0 = (i01 ... i0n1 )(i0n1 +1 ... i0n2 )...(i0nk +1 ... i0n ), where 1 ≤ n1 < ... < nk < n. Pick an h ∈ Sn such that h : ij 7−→ i0j , for all 1 ≤ j ≤ n. Then g 0 = h−1 gh, by (8.1). 2

8.3.2

Aside: Automorphisms of Sn

Though we shall not need it here, the following fact is interesting since it illustrates what a unique role the symmetric group S6 plays in the family of all symmetric groups. Theorem 149 If n 6= 2, 6 then the homomorphism φ : Sn → Aut(Sn ) (in part (c) of the exercises above) is an isomorphism: Sn ∼ = Aut(Sn ). If n = 6 then |Aut(S6 )| = 2 · |S6 |. Continuing our example from the appendix to the above chapter on the Platonic solids:

8.4. KERNELS AND NORMAL SUBGROUPS

161

Example 150 Any permutation of the 6 diagonals of the icosahedron, which may be regarded as an element of S6 , gives rise to a permutation of the set of 6 pentads, which may also be regarded as an element of S6 . This gives a map f : S6 → S6 , which is in fact a homomorphism. This homomorphism is injective so it is actually an automorphism. However, a 2-cycle on the set of 6 diagonals (i.e., swapping exactly 2 diagonals) does not induce a 2-cycle on the set of these 6 pentads. In fact, a 2-cycle on the set of diagonals gives rise to a product of three disjoint 2-cycles on the set of these 6 pentads. Therefore, by the above theorem (which says that an inner automorphism must preserve the cycle structure) this automorphism f cannot be an inner automorphism.

8.4

Kernels and normal subgroups

Let f : G1 → G2 be a homomorphism between two groups. Let ker(f ) = {g ∈ G1 | f (g) = e2 }, where e2 is the identity element of G2 . This set is called the kernel of f . Lemma 151 ker(f ) is a subgroup of G1 . Exercise 8.4.1 Prove this. Example 152 Let sgn : Sn → {±1} denote the homomorphism which associates to a permutation either 1, if it is even, or -1, if it is odd. Then An = ker(sgn) ⊂ Sn . The following properties of the kernel are useful: Lemma 153 : Let f : G1 → G2 be a homomorphism between two groups. (a) f is injective if and only if ker(f ) = {e1 }. (b) if g belongs to the kernel and x is any element of G1 then x−1 ∗ g ∗ x must also belong to the kernel.

162

CHAPTER 8. GROUPS, II

proof: (a) f is injective if and only if f (g1 ) = f (g2 ) implies g1 = g2 (g1 , g2 ∈ G1 ). Note f (g1 ) = f (g2 ) is true if and only if f (g1 ∗ g2−1 ) = e2 . If ker(f ) = {e1 } then f (g1 ∗ g2−1 ) = e2 implies g1 ∗ g2−1 = e2 , which implies g1 = g2 , which implies f is injective. Therefore, if ker(f ) = {e2 } then f is injective. Conversely, if f is injective then f (x) = f (e1 )(= e2 ) implies x = e1 (x ∈ G1 ). This implies ker(f ) = {e1 }. (b) Multiply both sides of e2 = f (g) on the left by f (x)−1 and on the right by f (x). We get e2 = f (x)−1 ∗ e2 ∗ f (x) = f (x−1 ) ∗ f (g) ∗ f (x) = f (x−1 ∗ g ∗ x), as desired. 2 Definition 154 Let H be a subgroup of G. We say that H is a normal subgroup if, for each g ∈ G, g −1 ∗ H ∗ g = H (i.e., for each g ∈ G and each h ∈ H, g −1 ∗ h ∗ g belongs to H). Notation: Sometimes we denote ”H is a normal subgroup of G” by H /G Example 155 An / Sn and |An | = 21 |Sn |. We have already shown the following Lemma 156 If f : G1 → G2 is a homomorphism between two groups then ker(f ) is a normal subgroup of G1 . Remark 15 The following remarkable result about the alternating group will not be needed in this course, except as an example of a group with no normal subgroups provided for the readers’ cultural benefit. It will not be proven here. (For a proof, see for example [R].) It is - believe it or not - connected with the fact that you cannot solve the general polynomial of degree 5 or higher using radicals, i.e., that there is no analog of the quadratic formula for polynomials of degree 5 or higher. Theorem 157 If X has 5 elements or greater then AX has no non-trivial proper normal subgroups. In other words, if H / AX is a normal subgroup then either H = {1} or H = AX .

8.5. QUOTIENT SUBGROUPS

163

The next fact about the alternating group will be needed later in our determination of the structure of the Rubik’s cube group. This fact also arose in connection with our previous discussion of the ”legal positions” of the 15 puzzle. Proposition 158 Let H be the subgroup of Sn generated by all the 3-cycles in Sn then H = An . proof: Since sgn : Sn → {±1} is a homomorphism, and since any 3-cycle is even, any product of 3-cycles must also be even. Therefore, H ⊂ An . If g ∈ An then g must swap an even number of the inequalities 1 < 2 < ... < n−1 < n, by Definition 37. Therefore, (since any permuation may be written as a product of 2-cycles, Theorem 58) g must be composed of permutations of the form (i j)(k l) or (i j)(j k). But (i j)(k l) = (i j k)(j k l) and (i j)(j k) = (i j k). Therefore, g ∈ H. This implies An ⊂ H, so An = H. 2 The following more precise result is very useful for the purposes of the analsis of permutation puzzles. Lemma 159 ([W]) Let X be a finite set, |X| ≥ 3 and fix u, v as elements in X. Then the 3-cycle (u, v, x), x an element of X − {u, v}, generates AX . This lemma proves an even stronger statement than the previous claim. Now, instead of only one element being fixed, there are two elements fixed and the group is still generated.

8.5

Quotient subgroups

One of the most useful facts about normal subgroups is the following: Lemma 160 If H is a normal subgroup of G then G/H is a group with the following operation: aH ∗ bH = (ab)H,

(aH)−1 = a−1 H,

for all a, b belonging to G. The identity element of this group is the trivial coset H. Exercise 8.5.1 Prove this.

164

CHAPTER 8. GROUPS, II

This group G/H is called the quotient group of G by H and is sometimes pronounced ”G mod H”. Example 161 If f : G1 → G2 is a homomorphism between two groups then G1 /ker(f ) is a group. The ”basic building blocks” of the collection of finite groups are the groups which have no non-trivial quotient groups. Intuitively, this is because a nontrivial quotient group is closely related to the original group but smaller in size (and hence perhaps subject to analysis by an inductive argument of some type). These ”basic” groups are called ”simple”: Definition 162 A simple group is a group with no proper normal subgroups other than the trivial subgroup {1}. Example 163 If p is a prime then Cp is simple. If n > 4 then An is simple (as was stated above in Theorem 157). These facts are proven in [R]. Simple groups are not very abundant. In fact, the first non-abelian simple group is of order 60 (its A5 ). The following basic result describes the quotient group G1 /ker(f ). Theorem 164 (”first isomorphism theorem”) If f : G1 → G2 is a homomorphism between two groups then G1 /ker(f ) is isomorphic to f (G1 ). proof: ker(f ) is a normal subgroup of G1 , so G1 /ker(f ) is a group. We must show that this group is isomorphic to the group f (G1 ). Define φ : G1 /ker(f ) → f (G2 ) by φ(g · ker(f )) = f (g), for g ∈ G1 . We must show (a) φ is well-defined, (b) φ is a homomorphism, (c) φ is a bijection. If g · ker(f ) = g 0 · ker(f ) then g −1 g 0 ∈ ker(f ), since ker(f ) is a group. This implies f (g −1 g 0 ) ∈ f (ker(f )) = {1}, so f (g) = f (g 0 ). This implies φ is well-defined. Since ker(f ) is normal, (g·ker(f ))(g 0 ·ker(f )) = gg 0 (g 0 −1 ker(f )g 0 )ker(f ) = gg 0 · ker(f ). Therefore φ((g · ker(f ))(g 0 · ker(f ))) = φ(gg 0 · ker(f )) = f (gg 0 ) = f (g)f (g 0 ) = φ(g · ker(f ))φ(g 0 · ker(f )), for all g, g 0 ∈ G. This implies φ is a homomorphism.

8.6. DIRECT PRODUCTS

165

It is clear that φ is surjective. To show that φ is a bijection, it suffices to prove φ is an injection. Suppose that φ(g · ker(f )) = φ(g 0 · ker(f )), for some g, g 0 ∈ G. Then f (g) = f (g 0 ), so f (g −1 g 0 ) = 1. By definition of the kernel, this implies g −1 g 0 ∈ ker(f ), so g · ker(f ) = g 0 · ker(f ). This implies φ is injective. 2 The other isomomorphism theorems will not be needed but will be stated to help to illustrate the usefulness of the notion of normality: Theorem 165 (”second isomorphism theorem”) If H, N are subgroups of a group G and if N is normal then (a) H ∩ N is normal in H, (b) there is an isomorphism H/(H ∩ N ) ∼ = N H/N. Theorem 166 (”third isomorphism theorem”) If N1 , N2 are subgroups of a group G, if N1 ⊂ N2 , and if N1 and N2 are both normal then (a) N2 /N1 is normal in G/N1 , (b) there is an isomorphism between (G/N1 )/(N2 /N1 ) ∼ = G/N 2. We shall not prove this result here - see [G] or [R].

8.6

Direct products

Definition 167 Let H1 , H2 be two subgroups. We say that a group G is the direct product of H1 with H2 , written G = H1 × H2 , if (a) G = H1 × H2 (Cartesian product, as sets), (b) the group operation on G is given ”coordinate-wise” (still denoted ”*” for simplicity): (x1 , y1 ) ∗ (x2 , y2 ) = (x1 ∗ x2 , y1 ∗ y2 ), for x1 , x2 ∈ H1 , y1 , y2 ∈ H2 (where ∗ denotes multiplication in H1 , H2 , and G).

166

CHAPTER 8. GROUPS, II

Example 168 Let G be (as a set) the Cartesian product G = C22 , where X 2 = X × X and where C2 denotes the cyclic group of order 2 (with addition mod 2 as the operation). Define addition on G coordinate-wise: (m1 , n1 ) + (m2 , n2 ) = (m1 + m2 , n1 + n2 ), where 0 ≤ mi ≤ 1, 0 ≤ nj ≤ 2, for i = 1, 2, j = 1, 2. Example 169 The symmetry group O of the octahedron is isomorphic to S4 × C2 . The symmetry group I of the icosahedron is isomorphic to A5 × C2 . (This is not isomorphic to S5 , despite the fact that they both have the same number of elements and they both contain A5 as a normal subgroup.) Example 170 Consider the subgroup H of the Rubik’s cube group generated by the ”square slice moves”, H =< MR2 , MD2 , MU2 > . Then H =< MR2 > × < MD2 > × < MU2 >∼ = C2 × C2 × C2 = C23 .

8.6.1

First fundamental theorem of cube theory

We shall frequently need the following fact: Theorem 171 Beginning with a solved cube, label the following facets with an invisible ”+” (i.e., mark the spatial position of the facet on the cube with a ”+”): • U facet of the uf edge subcube • U facet of the ur edge subcube • F facet of the fr edge subcube • all facets which can be obtained from these by a move of the slice group. Label the U and D facets of each corner subcube with an invisible ”+”. These ”+” signs are called the standard reference markings. Each move g of the Rubik’s cube yields a new collection of ”+” labels, called the markings relative to g. A position of the Rubik’s cube is determined by the following decision process: (a) How are the edge subcubes permuted?

8.6. DIRECT PRODUCTS

167

(b) How are the center subcubes permuted? (c) How are the corner subcubes permuted? (d) Which of the relative edge markings are flipped (relative to the standard reference markings)? (e) Which of the relative corner markings are rotated from the standard reference markings and, if so, by how much (2π/3 or 4π/3 radians clockwise, relative to the standard reference markings)? This is labeled as a theorem because of its relative importance for us in this course, not because of it’s difficulty! This is the ”First Fundamental Theorem” of ”mathematical cube theory”. Exercise 8.6.1 Prove this.

8.6.2

The twists and flips of the Rubik’s cube

We recall some notation: • X is the set of 48 facets of the Rubik’s cube which are not center facets, • V denotes the subset of facets which belong to some corner subcube, • E is the subset of facets which belong to some edge subcube. • Let G denote the Rubik’s cube group. • Let F be the group generated by all the moves of the Rubik’s cube group which do not permute any corner or edge subcubes but may twist or flip them. • Let SX , SV , SE , denote the symmetric group on X, V, E, respectively. We may regard F, G, as subgroups of SX . We may also regard SV , SE as subgroups of SX (for example, SV is the subgroup of SX which leaves all the elements of E fixed). • Let GV = SV ∩ G, GE = SE ∩ G, FV = SV ∩ F, FE = SE ∩ F.

168

CHAPTER 8. GROUPS, II

Note that the action of G on X induces an equivalence relation as follows: we say that a facet f1 is equivalent to a facet f2 if there is a move of the Rubik’s cube which sends one facet to the other. There are exactly two equivalence classes, or orbits, of G in X: namely, V and E. In particular, the action of G on V is transitive and the action of G on E is transitive. On the other hand, F leaves each vertex (resp., edge) fixed, though it may permute the corner facets (resp., edge facets) associated to a vertex (resp., edge). Exercise 8.6.2 Show that: (a) The set V of equivalence classes of F acting on V is in bijective correspondence with the set of all vertices of the cube. (b) The set E of equivalence classes of F acting on E is in bijective correspondence with the set of all edges of the cube. The interesting thing is that we have F = FV × FE

(direct product).

Exercise 8.6.3 Notation as above. (a) Show F = FV × FE (direct product). (b) Is SX = SV × SE (direct product)? A harder question, which we will answer later in the negative: Is G = GV × GE (direct product)?

8.6.3

The slice group of the Rubik’s cube

The material below can also be found in [BH]. Some ideas are also discussed in [Si]. Let H be the group < MR , MF , MU > generated by the middle slice moves. This group is called the ”slice group”. Let E be the set of edges of the cube (which we identify with the set of edge subcubes), let C be the set of center facets of the cube, and let X = E ∪ C. H acts on X. Note that H does not affect the corners (i.e, the vertices of the cube). Questions: (a) Is the action of H on X transitive? (b) Is the action of H on C transitive? (c) Is the action of H on E transitive?

8.6. DIRECT PRODUCTS

169

The answer to (a) is no - an edge subcube cannot be sent to a center facet, for example, so there is an element of X which cannot be sent to any other element of X by an element of H. The answer to (b) is yes - any center facet can be sent to any other center facet by an element of H. The answer to (c) is no - for example, the uf edge subcube cannot be sent to the ur edge subcube by a slice move, so there is an element of E which cannot be sent to any other element of E by an element of H. The answer of ”no” to (c) brings about the following Question: What are the orbits of H on E? The answer may be phrased in various ways, but let us look at it in the following way: suppose we call two edge subcubes equivalent if one can be sent to the other by a slice move (i.e., an element of H). There are 3 disjoint equivalence classes: all the subcubes in the middle RL-slice are equivalent, all the subcubes in the middle FB-slice are equivalent, and all the subcubes in the middle UD-slice are equivalent. The distinct orbits of H acting on E are the following: • the middle RL-slice, denoted by ERL , • the middle FB-slice, denoted by EF B , • the middle UD-slice, denoted by EU D . Note that E = ERL ∪ EF B ∪ EU D , is a partitioning of E into the distinct equivalence classes defined by the action of H on E. Each element of H determines an element in SX . We have a homomorphism f : H → SX This is another way of saying that H acts on the set X, which we already know. Note that each basic slice move M (so M is either MR , MF , or MU ) is, as an element of SX , of the following form: M = (4 − cycle in SE )(4 − cycle in SC ). Conversely, does an element of SX uniquely determine an element of H? In other words, is f injective (i.e., an embedding)?

170

CHAPTER 8. GROUPS, II

To answer this, fix an h ∈ H and think about what f (h) tells us: f (h) tells us which each subcube moves to which other subcube but it doesn’t tell us, for example, how a subcube is flipped or rotated. The fundamental theorem 171 insires the following question: Question: Can an element of H flip, but not permute, an edge subcube (and possibly permuting or flipping other subcubes of the cube)? The answer is no. The reason why is that the slice moves can only rotate a given edge subcube within the slice it belongs to. It follows, therefore, that the permutations of the edge subcube and centers determine a unique element of the slice group. In other words, we have proven the following Proposition 172 The homomorphism f : H → SX is an embedding. Remark 16 The analog of this for the Rubik’s cube group is false! H acts on the set ERL , so we have a homomorphism rRL : H → SERL and similarly, rU D : H → SEU D , rF B : H → SEF B . H acts on each of the sets E and C, so we have homomorphisms r = rRL × rU D × rF B : H → SERL × SEU D × SEF B ⊂ SE ,

s : H → SC ,

which we can put together to obtain an injective homomorphism r × s : H → SERL × SEF B × SEU D × SC To determine H, we determine the image of H in SERL × SEF B × SEU D × SC . To do this, we first look at the image of H in each of SERL , SEF B , and SEU D . This is easy enough: • the image of H in SERL is < MR >∼ = C4 , • the image of H in SEF B is < MF >∼ = C4 ,

8.6. DIRECT PRODUCTS

171

• the image of H in SEU D is < MU >∼ = C4 . Later, we shall want to think of C4 as {0, 1, 2, 3}, with addition mod 4, and the image of an element h ∈ H under one of the homomorphisms above, rRL : H → SEF B say, as an integer 0 ≤ rRL (h) ≤ 3. Next, we must determine the image of H in SC . This is easy if it’s looked at in the right way. As far as the movements of the center facets is concerned, the slice moves may be replaced by their corresponding rotations of the entire cube about an axis of symmetry. In this case, we see that the image of H in SC is the same as the image of the orientation-preserving symmetry group of the cube! This we know, by the discussion in chapter 7 and Example 144 above, is isomorphic to S4 . Putting all this together, we see that the image of H in SERL × SEF B × SEU D × SC is isomorphic to a subgroup of C43 × S4 . We may represent the elements of H , therefore, as 4-tuples (h1 , h2 , h3 , h4 ), with h1 , h2 , h3 ∈ C4 and h4 ∈ S4 . Since each of the generating moves of H (namely, MR , MU , and MF ) satisfies sgn(r(h)) = sgn(s(h)), for all h ∈ H, the image of H cannot be all of C43 × S4 . Proposition 173 The image of H in C43 × S4 . is isomorphic to the kernel of the map t : C43 × S4 → {±1} (h1 , h2 , h3 , h4 ) 7−→ sgn(h1 ) · sgn(h2 ) · sgn(h3 ) · sgn(h4 ), where each sgn is the sign of the permutation, regarded as an element of SX . Exercise 8.6.4 Show that |ker(t)| = (43 · 4!)/2 = 768. proof: We have shown that H is isomorphic to a subgroup of C43 × S4 . In fact, we know that the basic slice moves MR , MU , MF (which generate H) all belong to the kernel of t, so H is isomorphic to a subgroup of ker(t) < C43 × S4 .

172

CHAPTER 8. GROUPS, II

It remains to show that every element in ker(t) belongs to H. To do this, we consider the projection homomorphism p : H → S4 obtained by composing the homomorphism r × s : H → C43 × S4 constructed above with the projection homomorphism C43 × S4 → S4 . We have shown that p is surjective. Our next objective is to compute the kernel of p and use the first isomorphism theorem to determine H. Claim: The kernel of p is ker(p) = {h ∈ H | s(h) = 1, rRL (h) + rU D (h) + rF B (h) ≡ 0 (mod 2)}. Note that ker(p) is a subgroup of H so the sign of the permutation s(h) is equal to the sign of the permutation r(h): sgn(s(h)) = sgn(r(h)) = sgn(rRL (h)) · sgn(rU D (h)) · sgn(rF B (h)). This implies that ker(p) ⊂ {h ∈ H | s(h) = 1, rRL (h) + rU D (h) + rF B (h) ≡ 0 (mod 2)}. Conversely, pick an h ∈ H such that s(h) = 1 and rRL (h)+rU D (h)+rF B (h) ≡ 0 (mod 2). We may represent this element h as a 4-tuple (n1 , n2 , n3 , s), with 0 ≤ n1 , n2 , n3 ≤ 3 and s = 1 ∈ S4 . For example, • the element M1 = MR ∗ MF−1 ∗ MD ∗ MF is represented by the 4-tuple (1, 1, 0, 1), • the element M2 = MR ∗ MD ∗ MF ∗ MD−1 is represented by the 4-tuple (0, 1, 1, 1), • the element M3 = MF ∗ MD ∗ MR−1 ∗ MD−1 ∗ MF ∗ MD−1 ∗ MR ∗ MD is represented by the 4-tuple (0, 0, 2, 1). These elements generate all the elements of the group {(a, b, c, 1) | a, b, c ∈ C4 , a + b + c ≡ 0 (mod 2)}. Note that the group {(a, b, c) | a, b, c ∈ C4 , a + b + c ≡ 0 (mod 2)} is, in turn, the kernel of the map C43 → C2 given by (a, b, c) 7−→ a + b + c ≡ 0 (mod 2).

8.7. SEMI-DIRECT PRODUCTS

173

Exercise 8.6.5 Show that |{(a, b, c) | a, b, c ∈ C4 , a + b + c ≡ 0 (mod 2)}| = 43 /2 = 32. Therefore, the element h choosen above must be expressible as a ”word” in these three elements M1 , M2 , M3 . This shows that {h ∈ H | s(h) = 1, rRL (h) + rU D (h) + rF B (h) ≡ 0 (mod 2)} ⊂ ker(p), which implies the claim. To summarize what we have so far: we have a surjective homomorphism p : H → S4 with kernel {h ∈ H | s(h) = 1, rRL (h) + rU D (h) + rF B (h) ≡ 0 (mod 2)}. The kernel has |ker(p)| = 32 elements and the image has |im(p)| = |S4 | = 4! = 24 elements. Since, by the first isomomorphism theorem, H/ker(p) ∼ = S4 , we have |H| = 32 · 24 = 768. But the kernel of the homomorphism t : C43 × S4 → {±1}, which we know contains H as a subgroup, also has 768 elements. This forces h = ker(t). 2

8.7

Semi-direct products

If a group G contains two subgroups H1 and H2 , with H1 ¢ G normal, such that each element of G can be written uniquely as a product h1 h2 , with h1 ∈ H1 and h2 ∈ H2 then we say that G is the semi-direct product of H1 and H2 . In this situation, H2 is called a complement of H1 . In this section, we shall define two more ways of seeing how a semi-direct product can be expressed. Later, we shall see that the Rubik’s cube group is an easily described subgroup of a certain semi-direct product. Definition 174 Now suppose that H1 , H2 are both subgroups of a group G. We say that G is the semi-direct product of H1 by H2 , written G = H1 >¢ H2 if (a) G = H1 ∗ H2 , (b) H1 and H2 only have 1, the identity of G, in common, (c) H1 is normal in G. This is the ”internal version” of the semi-direct product.

174

CHAPTER 8. GROUPS, II

Of course, if we define anything using two apparently different definitions, we’d better be sure that they are equivalent! The fact that they are is a theorem (Theorem 7.23 in [R]) which we won’t prove here. Note that the multiplication rule in G doesn’t have to be mentioned since we are assuming here that G is given. The ”external version” is defined by a construction as follows: Definition 175 Assume we have a homomorphism φ : H2 → Aut(H1 ). Define multiplication on the set H1 × H2 by (x1 , y1 ) ∗ (x2 , y2 ) = (x1 ∗ φ(y1 )(x2 ), y1 ∗ y2 ). This defines a group operation. This group, denoted H1 >¢φ (H2 ), is the (external) semi-direct product. n This definition will be used with the example of H1 = Cm and H2 = Sn in the next chapter. These last two definitions are equivalent by Theorems 7.22-7.23 in [R]. As a set, H1 >¢ H2 is simply the Cartesian product H1 × H2 .

Example 176 Let R2 denote the direct product of the additive group of real numbers with itself: R2 = {(x, y) | x, y real}, with the group operation being addition performed componentwise. Let C2 denote the multiplicative cyclic group with 2 elements, whose elements we write (somwhat abstractly) as C2 = {1, s}. (We may think of s as being equal to -1 but there is a reason for this notation which shall be made clear soon.) Define an action of C2 on R2 by 1(x, y) = (x, y),

s(x, y) = (y, x),

(x, y) ∈ R2 .

Let G be the set G = R2 × C2 . Define the binary operation ∗ : G × G → G by (g1 , z1 ) ∗ (g2 , z2 ) = (g1 + z1 (g2 ), z1 ∗ z2 ),

8.7. SEMI-DIRECT PRODUCTS

175

for all g1 , g2 ∈ G and all z1 , z2 ∈ C2 . This is a group - the semi-direct product of R2 with C2 . To see this, we must answer some questions: (a) closed under the operation? Yes (b) existence if identity? Yes, e = ((0, 0), 1) (c) existence of inverse? Yes, ((x, y), 1)−1 = ((−x, −y), 1), and ((x, y), s)−1 = ((−y, −x), s). (d) associative? This is the hard one: ((g1 , z1 ) ∗ (g2 , z2 )) ∗ (g3 , z3 ) = (g1 + z1 (g2 ), z1 ∗ z2 ) ∗ (g3 , z3 ) = (g1 + z1 (g2 ) + (z1 ∗ z2 )(g3 ), (z1 ∗ z2 ) ∗ z3 ) (g1 , z1 ) ∗ ((g2 , z2 ) ∗ (g3 , z3 )) = (g1 , z1 ) ∗ (g2 + z2 (g3 ), z2 ∗ z3 ) = (g1 + z1 (g2 ) + z1 (z2 (g3 )), z1 ∗ (z2 ∗ z3 )) This implies associativity. Exercise 8.7.1 Let G be the semi-direct product constructed in the above example. Show that R2 (which we identify with the subgroup {((x, y), 1) | (x, y) ∈ R2 } of G) is a normal subgroup. Example 177 Let S3 = {1, s1 , s2 , s1 ∗s2 , s2 ∗s1 , s1 ∗s2 ∗s1 }, H1 = {1, s2 , s1 ∗s2 ∗s1 }, H2 = {1, s1 }. Let φ : H2 → Aut(H1 ) be defined by φ(1) = 1 (the identity automorphism) φ(s1 )(h) = s−1 1 ∗ h ∗ s1 = s1 ∗ h ∗ s1 (since s−1 1 = s1 ), h ∈ H1 . Define the (external) semi-direct product of H1 , H2 by G = H1 >¢ H2 . Exercise 8.7.2 Let G be the semi-direct product constructed in the above example. Show that H1 (which we identify with the subgroup {(h, 1) | h ∈ H1 } of G) is a normal subgroup. There is of course a close relationship between internally defined semidirect products and extenerally defined ones. The following lemma, which is proven in [R], explains this connection:

176

CHAPTER 8. GROUPS, II

Lemma 178 If G is the (internal) semi-direct product of H1 by H2 (so H1 is a normal subgroup of G) then there is a homomorphism φ : H2 → Aut(H1 ) such that G ∼ = H1 >¢φ (H2 ). Example 179 Let Cd be the cyclic group of order d, which we may regard as a set as Cr = {0, 1, ..., d}, with addition performed mod d. Let N = Cdn , which we regard as the group of ”n-vectors” with ”coefficients” in Cd . Let H = Sn be the symmetric group on n letters, i.e., the group of all permutations p : {1, 2, ..., n} → {1, 2, ..., n}. The group H acts on N by permuting the indices, i.e., the coordinates of the vectors. For p ∈ Sn , define p∗ : Cdn → Cdn by p∗ (v) = (vp−1 (1) , ..., vp−1 (n) ),

v = (v1 , ..., vn ) ∈ Cdn .

Now, for p, q ∈ Sn and v, w ∈ Cdn , define (p, v) ∗ (q, w) = (pq, w + q(v)). This defines a semi-direct product Cdn >¢ H. It is known that Cdn >¢ H is isomorphic to the group of all ”Cn -valued monomial matrices” (see [R], Exercise 7.33).

8.8

Wreath products

Let G1 be a group, let G2 be a group acting on a finite set X2 . Fix some 2 labeling of X2 as say X2 = {h1, h2, ..., hm }, where m = |X2 | and let GX 1 denote the direct product of G1 with itself m times, with the coordinates labeled by the elements of X2 . Definition 180 The wreath product of G1 , G2 is the group 2 G1 wr G2 = GX >¢ G2 1 2 where the action of G2 on GX 1 is via its action on X2 .

8.8. WREATH PRODUCTS

177

In particular, to each t ∈ G1 wr G2 there is a g2 ∈ G2 . We denote this projection by g2 = pr(t). Define the base of the wreath product by B = {t ∈ G1 wr G2 | pr(t) = 1}, 2 so B = GX 1 .

Example 181 Let Rn denote the direct product of the additive group of reals with itself n times. The group operation on Rn is componentwise addition. Let Sn denote the symmetric group. This acts on Rn by permuting coordinates: if r ∈ Sn is a permutation then define r(x1 , ..., xn ) = (xr(1) , ..., xr(n) ), for (x1 ..., xn ) ∈ Rn . This action respects the addition operation: r(x1 + y1 , ..., xn + yn ) = r(x1 , ..., xn ) + r(y1 , ..., yn ), (x1 ..., xn ), (y1 ..., yn ) ∈ Rn . (Incidently, it also preserves scalar multiplication: r(a ∗ (x1 , ..., xn )) = a ∗ r(x1 , ..., xn ), for (x1 ..., xn ) ∈ Rn , a ∈ R, so r defines an invertible linear transformation on Rn ; in fact, there is a homomorphism Sn → Aut(Rn ), where Aut(Rn ) denotes the group of all invertible linear transformations on Rn ; do you recognize this homomorphism? It has occurred previously in the notes...) Let G be the set G = Rn × Sn and define a binary operation ∗ : G × G → G by (g1 , p1 ) ∗ (g2 , p2 ) = (g1 + p1 (g2 ), p1 ∗ p2 ), for all g1 , g2 ∈ G and all p1 , p2 ∈ Sn . This is a group. (a) closed under the operation? (b) existence if identity? e = ((0, 0), 1) (c) existence of inverse? ((x, y), 1)−1 = ((−x, −y), 1), ((x, y), p)−1 = (−p−1 (x, y), p−1 ). (d) associative? This is the hard one: ((g1 , p1 ) ∗ (g2 , p2 )) ∗ (g3 , p3 ) = (g1 + p1 (g2 ), p1 ∗ p2 ) ∗ (g3 , p3 ) = (g1 + p1 (g2 ) + (p1 ∗ p2 )(g3 ), (p1 ∗ p2 ) ∗ p3 ) (g1 , p1 ) ∗ ((g2 , p2 ) ∗ (g3 , p3 )) = (g1 , p1 ) ∗ (g2 + p2 (g3 ), p2 ∗ p3 ) = (g1 + p1 (g2 ) + p1 (p2 (g3 )), p1 ∗ (p2 ∗ p3 ))

178

CHAPTER 8. GROUPS, II

This implies associativity. This group is the wreath product of R with Sn : G = R wr Sn , where R is the base. |X2 |

Lemma 182 (a) The base B, which is isomorphic to the direct product G1 is a normal subgroup of G1 wr G2 , (b) (G1 wr G2 )/B is isomorphic to G2 .

,

For this, we refer to chapter 8 of [NST] or to [R]. Example 183 Let H be the enlarged Rubik’s cube group of all legal and illegal moves of the 3 × 3 Rubik’s cube. In other words, in addition to the usual basic moves (namely, R, L, U, D, F, B), we allow you to take apart the cube and reassemble it (but we do not allow you to remove stickers from the facets). Let C3 denote the group of all rotations of a particular corner subcube by a 120 degree angle (actually, this group depends on the corner being rotated but since these groups are all isomorphic we drop the dependence from the notation). Let C2 denote the group of all flips of a particular edge subcube (rotations by a 180 degree angle). (Again, this group depends on the edge being flipped but since these groups are all isomorophic we drop the dependence from the notation). Then we shall prove later that H = (C3 wr SV ) × (C2 wr SE ), where V is the set of corner subcubes and E is the set of edge subcubes.

8.8.1

Application to order of elements in Cm wr Sn

In some cases, wreath products turn out to be relatively concrete and familiar groups. Let Pn (Cm ) denote the group of all n × n permutation matrices with entries in Cm . We begin with the following result: Theorem 184 There is an isomorphism between Cm wr Sn and the group n Pn (Cm ) which sends an element (~v , f ) ∈ Cm wr Sn , ~v = (v1 , v2 , ..., vn ) ∈ Cm and f ∈ Sn , to the matrix P (f )~v .

8.8. WREATH PRODUCTS

179

This follows from a result (see Theorem 197) which will be proven in the next chapter. It is also a special case of an exercise in [R]. In any case, it is n clear from this that an element (~v , f ) ∈ Cm wr Sn , ~v = (v1 , v2 , ..., vn ) ∈ Cm and f ∈ Sn , is order d only if the permutation matrix P (f ) is order d. Indeed, (~v , f )2 = (~v + f (~v ), f 2 ), (~v , f )3 = (~v + f (~v ) + f 2 (~v ), f 3 ), .. . (~v , f )k = (~v + ... + f k−1 (~v ), f k ). We conclude with the following classification of the elements of order d in the wreath product. n Lemma 185 An element (~v , f ) ∈ Cm wr Sn , ~v = (v1 , v2 , ..., vn ) ∈ Cm and d d−1 f ∈ Sn , is order d if and only if f = 1 and ~v + ... + f (~v ) = 0.

This result can, in principle, be used in conjunction with with the explicit determination of the Rubik’s cube group given later to determine all the elements of a given order. As an application, the elements of order 2 of the Rubik’s cube group will be given later.

180

CHAPTER 8. GROUPS, II

Chapter 9 The Rubik’s cube and the word problem Further details in the following background material may be found in [R], [MKS]. Definition 186 Given a list L of questions, a decision algorithm for L is a uniform set of unambiguous instructions which, when applied to any question in L gives the correct answer ”yes” or ”no” after a finite number of steps.

9.1

Background on free groups

Let X = {x1 , ..., xn } denote a set and X −1 a set disjoint from X whose −1 −1 elements we denote by {x−1 defines 1 , ..., xn }. Assume that the map x 7−→ x −1 1 0 a bijection X → X . It will be convenient to let x = x, xi = 1, where 1 is an element not belonging to X ∪ X −1 which we will call the identity element. A word on X is a sequence w = (a1 , a2 , ..., aN ), where N > 0 is some integer and each ai belongs to X ∪ X −1 ∪ {1}. The sequence of all 1’s is called the empty word. The inverse of the word w is the word −1 w−1 = (a−1 N , ..., a1 ). 181

182 CHAPTER 9. THE RUBIK’S CUBE AND THE WORD PROBLEM If ai = yiei , where ei is in {0, 1, −1} and yi ∈ X, then we shall write the word eN w as w = y1e1 ...yN . Example 187 Let X = {R, L, U, D, F, B}. The set of words on X are in a bijective correspondence with the set of sequences of basic moves you can make on the Rubik’s cube. eN on X reduced if either w is empty or if the We call a word w = y1e1 ...yN exponents ei are non-zero and if there are no x ∈ X with x, x−1 adjacent in w.

Definition 188 The free group Fn = FX on the generators x1 , ..., xn is the group of all reduced words on X. The proof that Fn is a group is not entirely easy (verifying the associativity property is perhaps the hardest part), see Theorem 11.1 in [R].

9.1.1

Length

eN is a reduced word (so ei ∈ {0, 1, −1}) If, in the notation above, w = y1e1 ...yN then we call N the length (or reduced length, to be more precise) of w. If G =< g1 , ..., gk >⊂ Sn is a finite permutation group generated by permutations g1 , ..., gk then we may still define the notion of length:

Definition 189 Suppose g ∈ G is not the identity, where G is a permutation group as above. Then g may be written eN g = y1e1 ...yN ,

where each yi ∈ {g1 , ..., gk } and where ei ∈ {0, 1, −1}. The number N and the sets {y1 , ..., yN }, {e1 , ..., eN } may not be unique for a given g but among all such possibilities there is at least one such that the value of N is minimum. We call this the length of g, denoted `(g). Let X t`(g) . PG (t) = g∈G

This is called the Poincar´e polynomial of G.

9.1. BACKGROUND ON FREE GROUPS

183

The length of g is the distance in the Cayley graph between the vertex g and the vertex 1. As was mentioned in the chapter on graph theory, the problem of determining the largest possible distance in the Cayley graph of the Rubik’s cube group is known as ”God’s algorithm” and is currently unsolved. Example 190 Let G = Sn with generators gi = (i, i + 1), i = 1, ..., n − 1. The Poincar´e polynomial is known: Πnk=1

tk+1 − 1 , t−1

by [Hum], §3.15. In case n = 2, this is t3 + 2t2 + 2t + 1 = (t + 1)(t2 + t + 1) Problem What is the longest possible length of an element of the Rubik’s cube group (with respect to the generators R, L, U, D, F, B)? Problem What is the Poincare polyomial of the Rubik’s cube group (with respect to R, L, U, D, F, B)?

9.1.2

Trees

We may represent the free group graphically as follows. We define the Cayley graph of Fn inductively: • draw a vertex for each element of X ∪ X −1 (these are the vertices, V1 say, for the words of length 1), • Suppose we are given that you have already drawn all the vertices for the words of length k − 1, Vk−1 let’s call them. For each x ∈ X ∪ X −1 and each v ∈ Vk−1 , draw a vertex for each word of length k obtained by multiplying v by x on the right, v ∗ x, and connect v and v ∗ x by an edge. There are infinitely many vertices, each of which has degree |X ∪ X −1 |. Moreover, this graph has no circuits or loops (i.e., no path of edges crosses back over onto itself). Such a graph is called a tree.

184 CHAPTER 9. THE RUBIK’S CUBE AND THE WORD PROBLEM Example 191 Let X = {R, L, U, D, F, B}. The elements of the free group FX correspond to the mechanically different sequences of basic moves you can make on the Rubik’s cube. Of course, different sequences of moves may yield the same position of the Rubik’s cube (e.g., R4 and 1 are the same position but sequence of moves used to attain them are distinct). There are infinitely many vertices of the Cayley graph of FX , each of which corresponds to a mechanically distinct move of the Rubik’s cube.

9.2

The word problem

There is a way to list all the elements of Fn , called the lexicographic ordering. We shall describe a way to list all the words in Fn as though they were in a dictionary. We shall give an algorithm for determining if a word w ∈ Fn occurs before a word w0 , in which case we write w < w0 . In the dictionary below, we shall, for example, distingish between the the identity 1 and the ”non-reduced” word x1 ∗ x−1 1 . The first element in this lexicographically ordered list is the word 1, the next 2n words are the words −1 x1 < x−1 1 < ... < xn < xn .

In general, we define y1 ..yM < z1 ...zN if either (a) M < N or (b) M = N and y1 < z1 or y1 = z1 and y2 < z2 or y1 = z1 and y2 = z2 and y3 < z3 or ... . List all the elements of Fn as Fn = {w1 , w2 , ...}, so w1 = 1, w2 = x1 , .... Let G be a subgroup of Fn or a permutation group G =< g1 , ..., gn >. If G is a permutation group then we regard a word wk as an element of G by substituting gi for each xi , 1 ≤ i ≤ n. Definition 192 Fix a g ∈ G. We say that G has a solvable word problem if there is a decision algorithm for the list L of questions of the form: Is wk = g in G? Claim: Fn has a solvable word problem. proof The following decision algorithm yields a solvable word problem.

9.3. GENERATORS AND RELATIONS

185

1. If w is a word not equal to 1 then underline the first occurance, if any, of the expression xi ∗ x−1 i . If no such expression occurs in w then go to step 3, otherwise go to step 2. 2. delete the expression xi ∗ x−1 from w and go to step 1. i 3. If w = 1 then stop and return ”yes”, otherwise, stop and return ”no”. 2 Claim: Each finite permutation group has a solvable word problem. Exercise 9.2.1 Why? Theorem 193 A decision algorithm for the word problem for the Rubik’s cube group with generators R, L, U, D, F, B is the same as an algorithm for solving the Rubik’s cube.

9.3

Generators and relations

Here’s a hypothetical situation: You and a friend each have a robot on the planet Pluto with a scrambled Rubik’s cube. You and your friend also have duplicate cubes, scrambled the same way as your robots. (We can call the robots R2 D2 and R2 B 2 if you wish:). These robots have manual dexterity but no preprogramming on how to solve the cube. Furthermore, pretend it is very expensive to program different moves, so you want to teach the robot the smallest number of separate moves that you can. On the other hand, the moves need not be basic moves (U , R, ..) since it costs the same to teach the robot the move R as the move R ∗ U 2 ∗ R−1 , for example. Your solution will be a ”word” in these taught moves. Again, to minimize the cost of transmission, you want the ”word” to be absolutely as short as possible. A prize of 1 million dollars has been set up to the first of you who can get their robot to solve its cube. In other words, we want to solve the word problem for the cube and we want to do it as efficiently as possible. Suppose we know we need n generators and we know that this is the smallest number. How do we make a ”word” as short as possible? To make a word in these generators as small as possible, we must know all the ”relations” between these generators so we can, if

186 CHAPTER 9. THE RUBIK’S CUBE AND THE WORD PROBLEM necessary, substitute them into the word and perform some cancelation. This is what this section is about. Let X be a finite set, say n = |X|. Let Y be a set of reduced words on X. Let R be the smallest normal subgroup of Fn containing Y . Since R is normal, the quotient Fn /R is a group. Definition 194 Let G be a group. We say that G has generators X and relations Y if G is isomorphic to Fn /R. A collection of generators and relations defining a group is called a presentation of the group. As a matter of notation, an element r ∈ R is written as an equation r = 1 in G. Remark 17 For those with a background in topology: Serre [Ser], §3.3, gives a topological interpretation of R as ”the fundamental group of the Cayley graph of G with respect to X”. Example 195 The cyclic group of order 3, C3 , has one generator x and one relation x3 = 1, so X = {x},

R = {(x3 )k | k ∈ Z} ⊂ F1 = {xk | k ∈ Z}.

Here the cosets of F1 /R are R, xR, x2 R. The set of these three cosets is closed under multiplication. For example, (xR)(x2 R) = x(Rx2 )R = xx2 RR = x3 R = R, so the inverse of the element xR is x2 R. More generally, Cn has presentation Cn = {x | xn = 1}. Exercise 9.3.1 By constructing moves of the Rubik’s cube of order 2, 3, 4, show that the Rubik’s cube ”contains” the subgroups C2 , C3 , C4 . Lemma 196 The group Cm × Cn is presented by Cm × Cn = {a, b | am = 1, bn = 1, ab = ba}. Remark 18 In general, if G is generated by g1 , ..., gm with relations Ri (g1 , ..., gm ) = 1 and if H is generated by h1 , ..., hn with relations Si (h1 , ..., hn ) = 1 then G × H is the group generated by the gi , hj , with relations Ri (g1 , ..., gm ) = 1, Si (h1 , ..., hn ) = 1, and gi hj = hj gi . We shall not prove this but refer to [MKS], Exercise 13, §4.1.

9.4. GENERATORS, RELATIONS FOR GROUPS OF ORDER < 26 187 Exercise 9.3.2 By constructing moves of the Rubik’s cube of order 2, 3 which commute show that the Rubik’s cube ”contains” the subgroups C2 × C3 ∼ = C6 .

We conclude this section with a table of all the finite groups of order less than or equal to 25 and their generators and relations.

9.4

Generators, relations for groups of order < 26

The following table was obtained from the tables in [TW]. Notation:

Cn = cyclic group of order n, Dn = dihedral group of order 2n = symmetry group of the regular n-gon, Q = quaternion group = {−1, 1, −i, i, −j, j, −k, k}, Sn = symmetric group of permutations of {1, 2, ..., n}, An = alternating group of even permutations of {1, 2, ..., n}, Fq = finite field with q elements (q=power of a prime), Z/nZ = integers modulo n.

188 CHAPTER 9. THE RUBIK’S CUBE AND THE WORD PROBLEM

Order

Group G

generators

relations

2 3 4 4

C2 C3 C4 C2 × C2

a a a a, b

5 6 6

C5 C6 = C2 × C3 S3

a a a, b

7 8 8

C7 C8 C2 × C4

a a a, b

8

C2 × C2 × C2

a, b, c

8

D4

a, b

8

Q

a, b

9 9

C9 C3 × C3

a a, b

10 10

C10 = C2 × C5 D5

a a, b

11 12 12

a a a, b

12

C11 C12 = C3 × C4 C2 × C6 = C2 × C2 × C3 D6

12

A4

a, b

12

Q6

a, b

a2 = 1 a3 = 1 a4 = 1 a2 = 1, b2 = 1, ab = ba a5 = 1 a6 = 1 3 a = 1, b2 = 1, aba = b a7 = 1 a8 = 1 a2 = 1, b4 = 1, ab = ba 2 a = 1, b2 = 1, c2 = 1, ab = ba, bc = cb, ac = ca a4 = 1, b2 = 1, aba = b 4 a = 1, b2 = a2 , aba = b a9 = 1 a3 = 1, b3 = 1, ab = ba a10 = 1 a5 = 1, b2 = 1, aba = b a11 = 1 a12 = 1 a2 = 1, b6 = 1, ab = ba 6 a = 1, b2 = 1, aba = b a2 = 1, b3 = 1, (ba)3 = 1 a6 = 1, b2 = a3 , aba = b

a, b

notes

G = A3 Klein 4-group Aut(G) = GL(2, F2 )

Aut(G) = G G = GL(2, F2 )

Aut(G) = GL(3, F2 )

Aut(G) = GL(2, F3 )

Aut(G) = G ”dicyclic”

9.4. GENERATORS, RELATIONS FOR GROUPS OF ORDER < 26 189

13 C13 14 C14 = C2 × C7 14 D7

a a a, b

15 C15 = C3 × C5 16 C16 16 C2 × C8

a a a, b

16

C4 × C4

a, b

16

C22 × C4

a, b, c

16

C24

a, b, c, d

16

D4 × C2

a, b, c

16 16

Q × C2

a, b, c a, b, c

16

a, b

16

a, b

16

a, b

16

a, b

16

D8

a, b

16

Q8

a, b

17 C17 18 C18 = C2 × C9

a a

a13 = 1 a14 = 1 a7 = 1, b2 = 1, Aut(G) = G aba = b a15 = 1 a16 = 1 2 a = 1, b8 = 1, ab = ba 4 a = 1, b4 = 1, Aut(G) = GL(2, Z/4Z) ab = ba a2 = 1, b2 = 1, c2 = 1, ab = ba, ac = ca, bc = cb a2 = 1, b2 = 1, Aut(G) = GL(4, F2 ) 2 2 c = 1, d = 1, ab = ba, ac = ca, ad = da, bc = cb, bd = db, cd = dc a4 = 1, b2 = 1, c2 = 1, aba = b, ac = ca, bc = cb Exercise 2 a = 1, b2 = 1, c2 = 1, abc = bca = cab a2 = 1, b2 = 1, (ab)2 = 1, (a−1 b)2 = 1 a4 = 1, b4 = 1, a semidirect product aba = b of C4 with C4 8 2 a = 1, b = 1, a semidirect product 5 ab = ba of C8 with C2 (C2 normal) 8 2 a = 1, b = 1, a semidirect product ab = ba3 of C8 with C2 (C2 normal) 8 2 a = 1, b = 1, a semidirect product aba = b of C8 with C2 (C2 normal) 8 2 4 a = 1, b = a , aba = b a17 = 1 a18 = 1

190 CHAPTER 9. THE RUBIK’S CUBE AND THE WORD PROBLEM

18

C3 × C6 = C3 × C3 × C2 18 S3 × C3 18

D9

18

a, b a, b, c a, b

a, b, c

19 20 20

C19 C20 = C4 × C5 C2 × C10

a a a, b

20

D10

a, b

20

Q10

a, b

20

a, b

21 21

C21 = C3 × C7

a a, b

22 22

C22 = C2 × C11 D11

a a, b

C23 C24 = C3 × C8 C2 × C12 = C2 × C3 × C4 24 C22 × C6

a a a, b

23 24 24

a, b, c

a3 = 1, b6 = 1, ab = ba 3 a = 1, b2 = 1, c3 = 1 aba = b, ac = ca, bc = cb a9 = 1, b2 = 1, a semidirect product aba = b of C9 with C2 (C2 normal), Aut(G) = G 3 3 2 a = 1, b = 1, c = 1 ab = ba, aca = c, bcb = c a19 = 1 a20 = 1 a2 = 1, b10 = 1, ab = ba 10 a = 1, b2 = 1, aba = b 10 a = 1, b2 = a5 , aba = b a5 = 1, b4 = 1, a semidirect product 3 ab = ba of C5 with C4 (C4 normal), Aut(G) = G 21 a =1 a7 = 1, b3 = 1, a semidirect product 4 ab = ba of C7 with C3 (C3 normal), Aut(G) = G 22 a =1 11 a = 1, b2 = 1, Aut(G) = G aba = b a23 = 1 a24 = 1 2 a = 1, b12 = 1, ab = ba 2 a = 1, b2 = 1, c6 = 1, ab = ba, ac = ca, bc = cb

9.4. GENERATORS, RELATIONS FOR GROUPS OF ORDER < 26 191 24

D6 × C2

a, b, c

24 24 24 24 24 24

A4 × C2 Q6 × C2 D4 × C3 Q × C3 S3 × C4 D12

a, b, c a, b, c a, b, c a, b, c a, b, c a, b

24

Q12

a, b

24

S4

a, b

24 SL(2, F3 ) a, b, c

a6 = 1, b2 = 1, c2 = 1, aba = b, ac = ca, bc = cb Exercise Exercise Exercise Exercise Exercise 12 a = 1, b2 = 1, aba = b a12 = 1, b2 = a6 , aba = b 4 a = 1, b2 = 1, (ab)3 = 1 a4 = 1, b2 = a2 , c3 = 1, aba = b, ac = cb, bc = cab

24

a, b

a3 = 1, b8 = 1, aba = b

24

a, b, c

a3 = 1, b4 = 1, c2 = 1, bcb = c, aba = b, ac = ca a25 = 1 Exercise

25 25

C25 C52

a a, b

Aut(G) = G Aut(G) = Aut(Q) = S4 , a semidirect product of Q with C3 (C3 normal) a semidirect product of C3 with C8 (C8 normal) a semidirect product of C3 with D4 (D4 normal)

Problem Which of these is (isomorphic to) a subgroup of the Rubik’s cube group?

Problem Of those which are subgroups, find moves associated to the generators given which satisfy the relations given.

192 CHAPTER 9. THE RUBIK’S CUBE AND THE WORD PROBLEM

graph of (n,# gps of order n)

9.5

The presentation problem

The following problem is unsolved: Problem (Singmaster [Si]): Let G be the Rubik’s cube group. Find a set of generators and relations for G of minimal cardinality (i.e., |X| + |Y | is of minimal cardinality). Problem : Find (a) a set of generators for G of minimal cardinality, (b) a set of relations for G of minimal cardinality, (c) an expression for each such generator as a word in the basic moves R, L, U, D, F, B. The part (a) is known: there are 2 elements which generate G [Si]. Part (b) is not known (though Dan Hoey’s post of Dec 17, 1995 to the cube-lover’s

N 9.6. A PRESENTATION FOR CM >¢ SN +1

193

list may describe the best known results [CL]; he suggests that G has a set X of 5 generators and a set Y of 44 relations such that the total length of all the reduced words in Y is 605).

9.6

n A presentation for Cm >¢ Sn+1

This section was written with Dennis Spellman. n We can identify the Cm >¢ Sn+1 with the group of (n + 1) × (n + 1) invertible monomial matrices g with coefficients in Cm having the following condition on the determinant: if we write g = p · d, where p is a permutation matrix and d is a diagonal matrix then det(d) = 1 (this determinant 1 condition is a condition corresponding to the ”conservation of twists” for the moves of the Rubik’s cube). n We may identify Cm with the subgroup

{(x1 , ..., xn+1 ) | x1 x2 ...xn+1 = 1, xi ∈ Cm } n and Cm >¢ Sn+1 as a subgroup of the wreath product Sn+1 wr Cm . n+1 Consider G = Cm >¢ Sn+1 . The group Sn+1 has presentation

Sn+1 =< a1 , ..., an | (ai aj )mij = 1, ∀1 ≤ i, j ≤ n >, where    3,

j = i ± 1, mij =  2, |i − j| > 1,  1, i = j. The following diagram may help to visualize the exponents mij in the case n = 4:

194 CHAPTER 9. THE RUBIK’S CUBE AND THE WORD PROBLEM

As a group of (n + 1) × (n + 1) monomial matrices, we identify ai with the permutation matrix,          si =        

1

0 .. .

...

0

1 0 1 1 0 1 0

...

... 0

         .       

1

If I is the (n + 1) × (n + 1) identity matrix and if Eij denotes the matrix which is 0 in every entry except the ij entry, which is 1, then si = I − Eii − Ei+1,i+1 + Ei,i+1 + Ei+1,i .

N 9.6. A PRESENTATION FOR CM >¢ SN +1

195

The group Cm has presentation Cm =< h | hm = 1 > . n The group Cm has presentation n Cm =< h1 , ..., hn | hm i = 1, hi hj = hj hi , ∀1 ≤ i, j ≤ n + 1 > . n We identify Cm with the Cartesian product

{(h1 (x1 ), h2 (x2 ), ..., hn (xn )) | xi ∈ Cm }, where hi (t) is the diagonal matrix hi (t) = I − Eii − Ei+1,i+1 + tEi,i + t−1 Ei+1,i+1 . There are the following identities between the si and the hj (t): si hj (t)s−1 |i − j| > 1, i = hj (t), −1 si hi (t)si = hi (t)−1 , si±1 hj (t)s−1 i±1 = hi (t)hi±1 (t). This in mind motivates the formulation of the following statement (which we shall prove in the next section): Theorem 197 n Cm >¢ Sn+1 =< a1 , ..., an , h1 , ..., hn |

9.6.1

(ai aj )mij = 1, ∀1 ≤ i, j ≤ n, m hi = 1, hi hj = hj hi , ∀1 ≤ i, j ≤ n ai hj a−1 |i − j| > 1, i = hj , −1 ai hi ai = h−1 i , ai±1 hj a−1 = h h i i±1 > i±1

A proof

Let P denote the group presented in the above theorem. The claim is, of n >¢ Sn+1 ∼ course, that Cm = P . There is a surjective homomorphism f : n P → Cm >¢ Sn+1 given by sending the generators to the generators. The problem is to which that this is injective. Let K = ker(f ), so n |P | = |P/K||K| ≥ |P/K| = |Cm >¢ Sn+1 | = mn (n + 1)!.

196 CHAPTER 9. THE RUBIK’S CUBE AND THE WORD PROBLEM Note H =< h1 , ..., hn | hm i = 1, hi hj = hj hi > < P is a normal subgroup of P since each ai sends a generator of H to a product of them or their inverses. n Also, note H ∼ . = Cm We claim that P/H ∼ = Sn+1 . From this it will follow that |P | = mn (n+1)!, proving that |K| = 1, as desired. To establish P/H ∼ = Sn+1 , we show that the presentation on P/H one gets from Theorem 2.1 in [MKS] is the same as that of Sn+1 . This is actually easy to see: If W (a1 , ..., an , h1 , ..., hn ) = 1 is a relation in the presentation, we must determine the word W (a1 , ..., an , h1 , ..., hn )H in P/H. Note that evey relation ”collapses” except to the relations (ai aj )mij = 1. These relations define the presentation for Sn+1 , as desired. 2

Chapter 10 The 2 × 2 and 3 × 3 cube groups In this chapter, we describe mathematically the moves of the 2 × 2 and 3 × 3 Rubik’s cubes.

10.1

Mathematical description of the 3 × 3 cube moves

In this section, we describe mathematically the moves of the 3 × 3 Rubik’s cube. As we will see, this will lead eventually to the description of the Rubik’s cube group as a subgroup of index 12 of a direct product of two wreath products.

10.1.1

Notation

First, orient all the corners and edges as in theorem 171. These are depicted as follows, except that we have replaced the ”+” in theorem 171 with a white square: 197

198

and

CHAPTER 10. THE 2 × 2 AND 3 × 3 CUBE GROUPS

10.1. MATHEMATICAL DESCRIPTION OF THE 3×3 CUBE MOVES199

Let G =< R, L, U, D, F, B > be the group of the 3×3 Rubik’s cube and let H be the ”enlarged” group generated by R, L, U, D, F, B and all the ”illegal” moves (where one is allowed to disassemble and reassemble the cube but not remove any facets). Let V denote the set of vertices of the cube (which we identity with the set of corner subcubes of the Rubik’s cube) and let ρ : H → SV denote the homomorphism which associates to each move of the Rubik’s cube the cooresponding permutation of the vertices. Let E denote the set of edges of the cube (which we identity with the set of edge subcubes of the Rubik’s cube) and let σ : H → SE denote the homomorphism which associates to each move of the Rubik’s cube the corresponding permutation of the edges.

200

10.1.2

CHAPTER 10. THE 2 × 2 AND 3 × 3 CUBE GROUPS

Corner orientations

Let v : H → C38 be the function which associates to each move g ∈ H the corresponding corner orientations. More precisely, let g ∈ H and say g moves corner i to corner j. Then vi (g) ∈ C3 is the orientation which the ith vertex gets sent to by g, where the vertices are labeled as in the diagram shown and where the orientation is the number of 120o clockwise twists required to turn the relative reference ”+” obtained by moving corner i to j using the move g into the standard reference ”+” on corner j. Example 198 We have X F U F*U U*F D B R L

~v (X) (2,0,0,1,1,0,0,2) (0,0,0,0,0,0,0,0) (2,0,0,1,1,0,0,2) (2,0,0,1,1,0,0,2) (0,0,0,0,0,0,0,0) (0,1,2,0,0,2,1,0) (1,2,0,0,2,1,0,0) (0,0,1,2,0,0,2,1)

Remark 19 The effect of a move g ∈ H on the corner orientations may also be regarded as a relabeling of the ”+” markings. Note that a move g ∈ H has two effects on the corners: (a) a permutation ρ(g) ∈ SV of the vertices, (b) a reorientation of the vertices moves in (a). In particular, for g, h ∈ H, the orientation ~v (gh) can only differ from v(g) in the coordinates corresponding to the vertices permuted by h. We shall now verify that the ”relative” orientation ~v (gh) − ~v (g) is the same as the orientation ~v (h), provided one takes into account the effect of g on the vertices: ~v (h) = ρ(g)(~v (gh) − ~v (g)), i.e., Lemma 199 ~v (gh) = ~v (g) + ρ(g)−1 (~v (h)). proof: The move gh orients the ith corner subcube by vi (gh) and permutes the vertices by ρ(gh), by definition. On the other hand, gh will first act by g then h. The move g will reorient the ith corner subcube by vi (g) and send the ith vertex to the ρ(g)(i)th vertex.

10.1. MATHEMATICAL DESCRIPTION OF THE 3×3 CUBE MOVES201 To study the subsequent effect of h on this, let us subtract ~v (g) from ~v (gh), so that we are back to our original orientation (we will add ~v (g) back in later). Call this position the modified cube for now. The move h first orients the j th corner subcube of the modified cube by vj (h) and permutes it to vertex ρ(h)(j). The ith subcube of the modified cube comes from (via g) the ρ(g)−1 (i)th subcube of the original cube. Thus the ith corner subcube of the modified cube is, by means of h, reoriented by vρ(g)−1 (i) (h). To this we must add in vi (g) to get the total effect of gh on the ith vertex of the original: vi (gh) = vi (g) + vρ(g)−1 (i) (h)), for each 1 ≤ i ≤ 8, which implies Lemma 199. 2

10.1.3

Edge orientations

Let w : H → C212 be the function which associates to each move g ∈ H the corresponding corner orientations. More precisely, let g ∈ H and say g moves corner i to corner j. Then wi (g) ∈ C2 is the orientation which the ith vertex gets sent to by g, where the vertices are labeled as in the diagram shown and where the orientation is the number of 180o flips required to turn the relative reference ”+” obtained by moving corner i to j using the move g into the standard reference ”+” on corner j. Example 200 We have X F U F*U U*F B D R L

w(X) ~ (1,0,0,0,0,0,0,0,1,0,0,0) (1,0,1,0,0,0,0,0,0,0,0,0) (1,0,1,0,1,0,0,0,1,0,0,0) (1,1,1,0,0,0,0,0,1,0,0,0) (0,0,0,0,0,0,1,1,0,0,0,0) (0,0,0,0,0,0,0,0,0,1,0,1) (0,1,0,0,0,1,1,0,0,1,0,0) (0,0,0,0,1,0,0,0,1,0,0,0)

Remark 20 The effect of a move g ∈ H on the edge orientations may also be regarded as a relabeling of the ”+” markings.

202

CHAPTER 10. THE 2 × 2 AND 3 × 3 CUBE GROUPS

Note that a move g ∈ H has two effects on the edges: (a) a permutation σ(g) ∈ SE of the edges, (b) a reorientation of the edges which were moved in (a). In particular, for g, h ∈ H, the orientation w(gh) ~ can only differ from w(g) ~ in the coordinates corresponding to the edges permuted by h. We shall now claim that w(gh) ~ = w(g) ~ + σ(g)−1 (w(h)), ~

(10.1)

i.e., that wi (gh) = wi (g) + wσ(g)−1 (i) (h), for each 1 ≤ i ≤ 12. The proof of this is so similar to the proof of Lemma 199 that we leave it to the student to modify its proof to verify (10.1).

10.1.4

The semi-direct product

Consider the following direct product of two semi-direct products: H 0 = (C38 >¢ SV ) × (C212 >¢ SE ). Remark 21 This may also we written in the notation of wreath products as the following direct product of two wreath products: H 0 = (SV wr C38 ) × (SE wr C212 ). As a set, we think of H as belonging to C38 ×SV ×C38 ×SV . If we represent elements h, h0 of H as h = (v, r, w, s), h0 = (v 0 , r0 , w0 , s0 ) ∈ C38 × SV × C38 × SV then the group operation will be given by h ∗ h0 = (v, r, w, s) ∗ (v 0 , r0 , w0 , s0 ) = (v + P (r)(v 0 ), rr0 , w + P (s)(w0 ), ss0 ). Consider the function ι : H → (C38 >¢ SV ) × (C212 >¢ SE ) g 7−→ (v(g), ρ(g), w(g), σ(g)). Proposition 201 ι is an isomorphism, H ∼ = H 0.

10.2. SECOND FUNDAMENTAL THEOREM OF CUBE THEORY

203

proof: Since (~v (g), ρ(g), w(g), ~ σ(g)) ∗ (~v (h), ρ(h), w(h), ~ σ(h)) = (~v (g) + P (ρ(g))(~v (h)), ρ(g)ρ(h), w(g) ~ + P (σ(g))(w(h)), ~ σ(g)σ(h)), the map ι is a homomorphism. Since any reorientation and permutation can be achieved by some illegal move, ι must be surjective. By theorem 171, the kernel of ι is trivial (this is just a fancy way of saying that if no subcube is permuted or reoriented then the cube doesn’t change!). 2

10.2

Second fundamental theorem of cube theory

First, some preliminaries. We identify, as in §10.1, each g ∈ G with a 4-tuple (~v (g), ρ(g), w(g), ~ σ(g)), where • ρ(g) is the corresponding permutation of the set of vertices V of the cube, • σ(g) is the corresponding permutation of the set of edges E of the cube, • v(g), w(g) are ”orientations” defined in §10.1. Remark 22 Let Sn denote the symmetric group on n letters and identify SV with S8 , SE with S12 . By example 137, we know that (a) ρ : G → S8 is a homomorphism, (b) σ : G → S12 is a homomorphism. Question: Given a 4-tuple (v, r, w, s), where r, s are permutations of the corners, resp. edges, as above and v ∈ C38 ,

w ∈ C212 ,

what conditions on r, s, v, w insure that it corresponds to a possible position of the Rubik’s cube? The following result is, according to [BCG], due to Ann Scott.

204

CHAPTER 10. THE 2 × 2 AND 3 × 3 CUBE GROUPS

Theorem 202 (Second fundamental theorem of cube theory) A 4-tuple (v, r, w, s) as above (r ∈ S8 , s ∈ S12 , v ∈ C38 , w ∈ C212 ) corresponds to a possible position of the Rubik’s cube if and only if (a) sgn(r) = sgn(s), (”equal parity as permutations”) (b) v1 + ... + v8 ≡ 0 (mod 3), (”conservation of total twists”) (c) w1 + ... + w12 ≡ 0 (mod 2), (”conservation of total flips”). proof: First we prove the ”only if” part. That is, we assume that (v, r, w, s) ∈ SV × SE × C38 × C212 represents a (legally obtained!) position of the Rubik’s cube. From this we want to prove (a)-(c). Let g ∈ G be the element which moves the Rubik’s cube from the solved position to the position associated to this 4-tuple. Then r = ρ(g) and s = σ(g). We know that g may be written as a word in the basic moves R, L, U, D, F, B, say g = X1 ...Xk , where each Xi is equal to one of the R, L, U, D, F, B. Observe that if X is any one of these basic moves then sgn(ρ(X)) = sgn(σ(X)). Since sgn, ρ, and σ are homomorphisms, it follows that sgn(r) = sgn(ρ(g)) =

k Y i=1

sgn(ρ(Xi )) =

k Y

sgn(σ(Xi )) = sgn(σ(X)) = sgn(s).

i=1

This proves (a). We have verified (b) for the basic moves in example 198 above. Note that (i) the conservation of twists condition in (b) is true for (v1 , ..., v8 ) if and only if it is true for any permutation P (p)(v) = (v(1)p , ..., v(8)p ), (ii) if (v1 , .., v8 ) and (v10 , ..., v80 ) each satisfy the conservation of twists condition in (b) then their sum also satisfies it. As above, write g as a word in the basic moves R, L, U, D, F, B, say g = X1 ...Xk , where each Xi is equal to one of the R, L, U, D, F, B. We assume that this expression is minimal in the sense that we choose the Xi so that k is as small as possible. This k is called the length of g. (This length is the same as the distance from g to the identity in the Cayley graph of G.) We now prove (b) by induction on the length. We have already checked it for all words of length k = 1. Assume k > 1. By the formula giving the orientation of the product of two moves in terms of the two orientations of the moves, we have ~v (X1 ...Xk−1 Xk ) = ρ(X1 ...Xk−1 )−1 (~v (Xk )) + ~v (X1 ...Xk−1 ).

10.2. SECOND FUNDAMENTAL THEOREM OF CUBE THEORY

205

The term ρ(X1 ...Xk−1 )−1 (~v (Xk )) satisfies the conservation of twists condition in (b) by (i) above. The term ~v (X1 ...Xk ) satisfies the conservation of twists condition in (b) by the induction hypothesis. Their sum satisfies the conservation of twists condition in (b) by (ii) above. This proves (b). The proof of (c) is very similar to the proof of (b), except that we use example 200 in place of example 198. Exercise 10.2.1 Provide the details. Now, we must prove the theorem in the ”if” direction. In other words, assuming (a), (b), and (c) we must show that there is a corresponding legal position of the Rubik’s cube. This part of the proof is constructive. First, we prove a special case. Assume that r and s are both the identity and that (w1 , ..., w12 ) = (0, ..., 0). There is a move which twists exactly two corners and preserves the orientations and positions of all other subcubes. For example, the move g = (R−1 D2 RB −1 U 2 B)2 twists the ufr corner by 120o clockwise, the bdl corner by 240o clockwise, and preserves the orientations and positions of all other subcubes. This move can be easily modified, by a suitable conjugation, to obtain a move which twists any pair of corners, and preserves the orientations and positions of all other subcubes. These moves generate all possible 8-tuples satisfying the conservation of twists condition in (b). This proves the ”if” part of the theorem in the case that r and s are both the identity and that (w1 , ..., w12 ) = (0, ..., 0). Next, we prove another special case. Assume that r and s are both the identity and that (v1 , ..., v8 ) = (0, ..., 0). There is a move which flips exactly two edges and preserves the orientations and positions of all other subcubes. For example, the move g = LF R−1 F −1 L−1 U 2 RU RU −1 R2 U 2 R (found in [B], page 112) flips the uf edge, the ur edge, and preserves the orientations and positions of all other subcubes. This move can be easily modified, by a suitable conjugation, to obtain a move which flips any pair of edges, and preserves the orientations and positions of all other subcubes. These moves generate all possible 12-tuples satisfying the conservation of flips condition in (c). This proves the ”if” part of the theorem in the case that r and s are both the identity and that (v1 , ..., v8 ) = (0, ..., 0). As a consequence of these last two special cases, it follows that the ”if” part of the theorem is true in the case that r and s are both the identity.

206

CHAPTER 10. THE 2 × 2 AND 3 × 3 CUBE GROUPS

Finally, we prove our last special case. Assume that (v1 , ..., v8 ) = (0, ..., 0) and that (w1 , ..., w12 ) = (0, ..., 0). Consider the following three claims. • Given any three edges subcubes, there is a move which is a 3-cycle on these edges and preserves the orientations and positions of all other subcubes. • Given any three corners, there is a move which is a 3-cycle on these corners and preserves the orientations and positions of all other subcubes. • Given any pair of edges and any pair of corners, there is a move which is a 2-cycle on these edges, a 2-cycle on these corners, and preserves the orientations and positions of all other subcubes. Exercise 10.2.2 Verify these three claims. By proposition 158, we know that AE is generated by the edge 3-cycles above and that AV is generated by the corner 3-cycles above. In other words, we can construct a position of the Rubik’s cube associated to any 4-tuple (r, s, 0, 0), provided r ∈ AV and s ∈ AE . The subgroup AE × AV is index 4 in SE × SV since |Sn /An | = 2. The third type of move, the edge-corner 2-cycles above, does not correspond to an element of the subset AE × AV of the Rubik’s cube group because an edge 2-cycle is an odd permutation of the edges. Therefore, if we consider the subgroup of SE × SV generated by all three types of moves we will obtain either all of SE × SV or some subgroup of index 2 which properly contains AE × AV . The first possibility can be ruled out since it contradicts the parity condition in (a). The only subgroup of SE × SV of index 2 which properly contains AE × AV is the subgroup of elements satisfying the parity condition in (a). It follows that the ”if” part of the theorem is true in the case that v and w are both zero. The theorem is a consequence of these special cases because of the following Claim: There is always a move, no matter what position of the Rubik’s cube is in, which does not permute any subcubes but ”solves” the orientation of the cube so that v and w are both zero. Exercise 10.2.3 Prove this claim.

10.2. SECOND FUNDAMENTAL THEOREM OF CUBE THEORY

207

2 Corollary 203 G = {g = (v, r, w, s) ∈ H | (a), (b), (c) in the above theorem hold}.

10.2.1

Some consequences

We shall now reformulate the above fact about the Rubik’s cube group from a point of view which (to me anyway) allows us to count the number of elements it has easier. Let G0 = {(v, r, w, s) | r ∈ S8 , s ∈ S12 , v = (v1 , v2 , ..., v8 ), vi ∈ {0, 1, 2}, v1 + ... + v8 ≡ 0 (mod 3), w = (w1 , w2 , ..., w12 ), wi ∈ {0, 1}, w1 + ... + w12 ≡ 0 (mod 2)}. Define a binary operation ∗ : G0 × G0 → G0 by (v, r, w, s) ∗ (v 0 , r0 , w0 , s0 ) = (v + P (r)(v 0 ), r ∗ r0 , w + P (s)(w0 ), s ∗ s0 ). This defines a group structure on G0 . This is a subgroup of the enlarged Rubik’s cube group of index 6. Theorem 204 There is an isomorphism G0 ∼ = (C37 >¢ S8 ) × (C211 >¢ S12 ), where Cn is the cyclic group with n elements and >¢ denotes the semi-direct product and where Cnk (n = 2, 3, k = 7, 11) is identified with the subgroup of Cnk+1 defined by {v = (v1 , v2 , ..., vk ) | vi ∈ {0, 1, n − 1}, v1 + ... + vk ≡ 0 (mod n)}. In particular, |G0 | = |S8 ||S12 ||C211 ||C37 | = 8! · 12! · 211 · 37 . Theorem 205 The Rubik’s cube group G is the kernel of the homomorphism φ : G0 → {1, −1} (v, r, w, s) 7−→ sgn(r)sgn(s). In particular, G < G0 is normal of index 2 and |G| = 8! · 12! · 210 · 37 .

208

CHAPTER 10. THE 2 × 2 AND 3 × 3 CUBE GROUPS

Recall that the commutator subgroup G1 of G is the subgroup consisting of all finite products of commutators [g, h] = g ∗ h ∗ g −1 ∗ h−1 , where g, h are arbitary elements of G. Theorem 206 |G1 | = |G|/2. In fact, we can explicitly determine G1 . Theorem 207 G1 = {g ∈ G | sgn(ρ(g)) = sgn(σ(g)) = 1}. This basically follows from the fact that the commutator subgroup of Sn is An , for n > 4 (see chapter 3 of [R]; in fact, for n > 4, An is the only proper non-trivial normal subgroup of Sn ). The above theorem implies that |G/G1 | = 2 (to see this, use the first homomorphism theorem). From this, it clearly follows (from those who see it clearly) that G1 is a normal subgroup of G.

10.3

Rubika esoterica

This section is a brief summary of various facts about the Rubik’s cube group. The material has been taken from [B] and [BH]. Let G < S54 be the group of moves of the Rubik’s cube. 1. |G| = 227 · 314 · 53 · 72 · 11 = (4.3...) · 1019 . 2. G is generated (as a permutation group) by m991 = U ∗ B ∗ L ∗ U ∗ L−1 ∗ U −1 ∗ B −1 and m992 = R2 ∗ F ∗ L ∗ D−1 ∗ R−1 (using the notation of [B]).

10.3. RUBIKA ESOTERICA

209

3. The center of G is given by Z(G) = {1, m490 } where m490 is the superflip which leaves all corners alone and flips every edge: m490 = R ∗ L ∗ F ∗ B ∗ U ∗ D ∗ R ∗ L ∗ F ∗ B ∗ U ∗ F 2 ∗ MR ∗ ∗F 2 ∗ U −1 ∗ MR2 ∗ B 2 ∗ MR−1 ∗ B 2 ∗ U ∗ MR2 ∗ D 4.

• Every group H of order less than 13 is isomorphic to a subgroup of G. • Every non-abelian group H of order less than 26 is isomorphic to a subgroup of G. • C13 (the cyclic group of order 13) is not isomorphic to a subgroup of G. • D26 (the dihedral group of order 26) is not isomorphic to a subgroup of G.

5. Let Q denote the quaternion group: Q = {1, −1, i, −i, j, −j, k, −k}, where i2 = j 2 = k 2 = −1, ij = k, jk = i, ki = j, and in general, xy = −yx for x, y belonging to i, j, k. Let Q∗ =< m706 , m710 > < G where m706 = F 2 ∗ MR ∗ U −1 ∗ MR−1 ∗ U −1 ∗ MR ∗ U ∗ MR−1 ∗ U ∗ F 2 , m710 = F ∗ U 2 ∗ F −1 ∗ U −1 ∗ L−1 ∗ B −1 ∗ U 2 ∗ B ∗ U ∗ L. Then φ : Q → Q∗ is an isomorphism, where φ(i) = m706 and φ(k) = m710 .

210

10.3.1

CHAPTER 10. THE 2 × 2 AND 3 × 3 CUBE GROUPS

Coxeter groups

We shall not define what a Coxeter group is (see [Hum]) but instead state a general result from the theory of Coxeter groups, give an example, and then apply it to the Rubik’s cube. This section assumes much more of the reader than the previous sections. The hyperoctahedral group of rank 12 is the group W C12 of all 12 × 12 signed permutation matrices. This group may be roughly thought of as follows: consider two columns of 12 boxes, aligned next two each other - i.e., a column of 12 shoe boxes, with a separate (right/left) compartment for each shoe. The group W C12 is the group of all permutations of the 24 shoes which does not separate any shoe from its partner but may swap a left shoe with the right shoe in a shoe box. Another way to understand the group W C12 is by means of its isomorphism with the semi-direct product of S12 with C212 , where S12 acts on C212 by permuting the coordinates. This group is (as we have seen already) isomorphic to the subgroup of the enlarged Rubik’s cube group H (of legal and illegal moves) which leaves all the corner subcubes alone (i.e., does not permute or twist any corners). In general, the group W Cn of all n × n signed permutation matrices is a ”Coxeter group” and there is an extensive and well-developed theory of such groups [Hum]. For example, their Poincar´e polynomial is known (the Poincar´e polynomial was defined in section 9.1.1). We shall state one such result and then see how it applies to the case of the Rubik’s cube. The Poincar´e polynomial of W Cn is, by [Hum], §3.7, P (t) = Πni=1

t2k − 1 . t−1

For example, the Poincar´e polynomial of W C12 is P (t) = (t + 1)(t3 + t2 + t + 1)...(t23 + t22 + ... + t + 1). An an application of this fact, we conclude that there is an element of W C12 of order 24. This implies that there is an element of order 24 in the subgroup of the enlarged Rubik’s cube group H which leaves all the corner subcubes alone (i.e., does not permute or twist any corners).

10.3. RUBIKA ESOTERICA

10.3.2

211

The moves of order 2

In this section we present, as promised, a method for determining the elements of a given order of the Rubik’s cube group. We shall only compute the number of elements of order 2, though the method should, in principle, work more generally. Recall from §8.8.1, we know that an element (~v , r) ∈ Cm wr Sn , ~v = n (v1 , v2 , ..., vn ) ∈ Cm and r ∈ Sn , is order d if and only if rd = 1 and ~v + ... + rd−1 (~v ) = ~0. In particular, the elements of order 2 in C37 >¢ S8 are those (~v , r) with r of order 2 (hence must be a product of disjoint 2-cycles) and ~v ∈ C38 satisfying v1 + ... + v8 = 0 and vi + v(i)r = 0. We therefore may use the ”multiplication principle” of counting (see §2.4) to compute the number of elements of order 2 in C37 >¢ S8 to be 1 4!

Ã

8 2



6 2



4 2

!

1 3+ 3! 3

Ã

8 2



6 2



4 2

!

1 3+ 2! 2

Ã

8 2



6 2

!

Ã

3+

8 2

!

Exercise 10.3.1 (hard) Show that the number of elements of order 2 in C211 >¢ S12 is 15353184. The theorem above on the Rubik’s cube group implies: An element (~v , r, w, ~ s) of the enlarged Rubik’s cube group is order 2 if and only if it is an element of the Rubik’s cube group of order 2 and v1 + ...v8 = 0 (mod 3) and w1 + ... + w12 = 0 (mod 2). But the enlarged Rubik’s cube group is a direct product of groups, say H = H1 × H2 , so an element (h1 , h2 ) ∈ H is order 2 if and only if one of the following mutually exclusive cases hold: • h1 = 1, h2 6= 1 and h22 = 1, • h1 6= 1, h2 = 1 and h21 = 1, • h1 6= 1, h2 6= 1 and h21 = h22 = 1. This implies Lemma 208 The number of elements of order 2 of the Rubik’s cube group is 7273 ∗ 15353184 + 15353184 + 7273 = 111679067689 = (1.11...) ∗ 1011 .

= 7273.

212

10.4

CHAPTER 10. THE 2 × 2 AND 3 × 3 CUBE GROUPS

Mathematical description of the 2 × 2 cube moves

This section, which is based on [DL], derives the group structure of the 2 × 2 Rubik’s cube. A position on the 2 × 2 cube is determined by (a) a permutation of the vertices, and (b) the orientation of the corner sub-cubes. An illegal move on the 2 × 2 cube is a reassembly of the corners. Let H =< R, L, U, D, F, B, and all the illegal moves > . This will be called the enlarged 2 × 2 cube group. Let G =< R, L, U, D, F, B >. G is contained in H with G < H. Let C38 = {0, 1, 2}8 be the group of 8-tuples with coordinate-wise addition mod 3. Let v : H → C38 be defined as follows: Assume h ∈ H sends the ith corner to the j th corner. vi (h) is the number in C3 = {0, 1, 2} which describes the orientation that the standard reference marking of the ith corner is sent to relative to the standard reference marking of the j th corner. The values of v are tabulated in (198). Let SV be the group of permutation of corner sub-cubes. We may identify SV with S8 since we have labeled the corners 1, ..., 8. H is a subset of the Cartesian product SV × C38 . Let p(h) denote the permutation of the vertices of the cube associated to h ∈ H. We have (v, r) ∗ (v 0 , r0 ) = (v + r(v 0 ), r ∗ r0 ) (~v (g), p(g)) ∗ (~v (h), p(h)) = (~v (g) + p(g)~v (h), p(g) ∗ p(h)) = (~v (g ∗ h), p(g ∗ h)). It is not hard to show, based on the results of the previous section, that H = C38 >¢ S8 = {(v, r) | r ∈ SV , v ∈ C38 }. In other words, H is the wreath product of S8 and C3 . Theorem 209 A two-tuple (v, r) ∈ C38 × SV corresponds to a legal position iff v1 + ... + v8 ≡ 0 (mod 3) (conservation of twists). proof: PART 1: In this part, we show that any pair (v, r) as in the theorem (where v satisfies conservation of twists) corresponds to a legal move g in such a way that r = ρ(g) and v = ~v (g).

10.4. MATHEMATICAL DESCRIPTION OF THE 2×2 CUBE MOVES213 Case 1: Assume r = 1 and v is arbitrary. From the solved position, any two corners, corner i and corner j say, can be twisted so that corner i has orientation 1, corner j has orientation 2, and all other corners have orientation 0. Call such a move ei,j . Example: (R−1 ∗ D2 ∗ R ∗ B −1 ∗ U 2 ∗ B)2 is e2,7 . Let y = a1 ∗ e1,8 + ... + a7 ∗ e7,8 , where ai ∈ {0, 1, 2}. This is a move of the 2×2 Rubik’s cube of the form (v, 1) - in other words, it permutes nothing but may twist some corners. By construction, all moves of this form are legal. For each ai ∗ ei,8 there are three different possible positions (independent of all other aj ∗ ej,8 ). Since there three choices for each ai , there are a total of 37 distinct moves of the form y as above. On the other hand, there are exactly 37 possible moves of the form (v, 1) which satisfy the conservation of twists. (proof: If v = (v1 , ..., v8 ), vi ∈ {0, 1, 2}, and v1 + ... + v8 ≡ 0 (mod 3), then there are 3 ways to choose each of v1 , .., v7 but then once these are fixed the consevation of twists condition leaves no choice for v8 . This leaves a total of 37 choices.) These 37 possible moves include, of course, the legal moves of the form y above. Thus every move of the form (1, v), with v satisfying conservation of twists, is legal. Case 2: Assume v = ~0 and r is arbitrary. Recall S8 is generated by the two-cycles (see chapter 3 above, §§3.3-3.4). Claim 1: Given any pair of corners, there is a 2-cycle move which swaps them. (Example: F −1 ∗ U ∗ B ∗ U −1 ∗ F ∗ U 2 ∗ B −1 ∗ U ∗ B ∗ U 2 ∗ B −1 ). Once two corners have been swapped, you may correct the orientation of any sub-cube by Case 1. Thus any permutation which preserves orientations is a legal moves. Case 3: Assume v and r are both arbitrary but satisfying conservation of twists. By case 2, we may make a legal move that changes (v, r) to (v, 1). By case 1, (v, 1) is a legal move. PART 2: In this part, we show that any legal move satisfies conservation of twists. Assume (v, r) ∈ C38 × SV is a legal move. Define the length of a move g ∈ G to be the smallest number n of generators needed to create the move, written length(g) = n. Induction hypothesis: If a move is length n, it satisfies conservation of twists. step n = 1: Every ~v (x) where x ∈ {R, L, U, D, F, B} satisfies the conservation of twists. step n > 1: Assume the induction hypothesis is true for all lengths ≤ n−1.

214

CHAPTER 10. THE 2 × 2 AND 3 × 3 CUBE GROUPS

Let x be length n and write x = x1 ∗ x2 , where length(x1 ) ≤ n − 1 and length(x2 ) = 1. Then ~v (x) = ~v (x1 ) + p(x1 )~v (x2 ), by the group operation. Furthermore, v(x1 ) satisfies conservation of twists by the induction hypothesis. Since p(x1 ) simply permutes the coordinates of ~v (x2 ), ~v (x2 ) still satisfies the conservation of twists. The sum of moves satisfying conservation of twists still satisfies conservation of twists. Conclusion: by induction, any move (v, r) ∈ C38 × SV that is a legal move satisfies the conservation of twists. 2

Chapter 11 Other Rubik-like puzzle groups ”An expert is someone who knows some of the worst mistakes that can be made in his subject, and how to avoid them.” Heisenberg, Werner PHYSICS AND BEYOND, 1971 This chapter shall survey, sometimes without proofs, some results on the group-theoretical structure of some of the permutation puzzle groups, as discussed in [GT], [Lu], [B], chapter 2, [NST], chapter 19.

11.1

On the group structure of the skewb

This section is based on G. Gomes and J. Montague [GM]. Notation: We fix an orientation of the cube and label the sides by R, L, U, D, F, B as in the case of the Rubik’s cube. The 120 degree clockwise rotation of a corner is denoted by a 3-letter juxtaposition of the letters abbreviating the 3 faces which the corner meets. (When you twist a corner of the skewb you must permute three other corners but the opposite side of the skewb is unaffected.) Such a move will be called a basic move - there are 8 of them, though twisting about a corner and twisting about the antipodal opposite corner is basically the same move (up to a rotation of the entire cube.) For example, FRU denotes the 120 degree clockwise rotation of the front-right-up corner, leaving the rest of the cube alone. Let C denote the set of square center facets and V the set of vertices of the cube. 215

216

CHAPTER 11. OTHER RUBIK-LIKE PUZZLE GROUPS Let G =< F RU, F LU, BRU, BLU, DF R, DF L, BDR, BDL >

denote the group of all (legal) skewb moves. Let G∗ denote the group of all legal and ”illegal moves” (where disassembly then reassembly is allowed). On each square center facet of the skewb we may choose a vertex with the following property: if we draw an arrow pointing from the chosen vertex of the square to the diametically opposite vertex on the square then the moves of the skewb permute these arrows amongst themselves, except that some arrows may possibly be reversed. This determines an orientation of each center facet. Call this puzzle the super skewb. For this new puzzle, let Gsuper =< F RU, F LU, BRU, BLU, DF R, DF L, BDR, BDL > denote the group of all (legal) super skewb moves. Let G∗super denote the group of all legal and ”illegal moves” (where disassembly is allowed). We orient the corners as in the case for the Rubik’s cube. Let y(g) ∈ 8 C3 = {0, 1, 2}8 denote the orientation for the corners. For the superskewb, we orient the center facets similarly. Let z(g) ∈ C46 = {0, 1, 2, 3}6 denote the orientation for the centers. Let SC denote the symmetric group on the set C, SV the symmetric group on the set V . Claim: There are homomorphisms ρ : G → SC ,

σ : G → SV ,

given, for each move g ∈ G, by ρ(g) = permutation of the center facets associated to g, and σ(g) = permutation of the vertices associated to g. Let H = C38 × SC × SV and define ∗ : H × H → H by (y, r, s) ∗ (y 0 , r0 , s0 ) = (rr0 , ss0 , r−1 (y 0 ) + y).

11.1. ON THE GROUP STRUCTURE OF THE SKEWB

217

Let Hsuper = C38 × SV × C46 × SC and define ∗ : Hsuper × Hsuper → Hsuper by (y, r, z, s) ∗ (y 0 , r0 , z 0 , s0 ) = (r−1 (y 0 ) + y, rr0 , s−1 (z 0 ) + z, ss0 ). Observation: There is an embedding of G into H and an embedding of Gsuper into Hsuper . Let GC be the group that acts only on the center facets of the skewb, and GV the group that acts only on the vertices. Now, G = GC × GV . Every generator of G is a 3-cycle on the center facets. This means that r is an element of AC . It is a fact that the elements (i, j, k) of Sn generate An (for i, j, k elements of {1, 2, ..., n}). Therefore, GC = A6 . The group that acts on the vertices of the skewb is slightly more complicated. Unlike the Rubik’s cube, there is no condition like conservation of twists which applies to the entire vertex set. Instead, we must split the vertices of the skewb into two 4-corner orbits. This idea is borrowed from Bandelow’s notes on Mickey’s Challenge (a puzzle similar to the Masterball). An orbit is constructed by starting with one corner and including the opposite corner of each face that meets at the first corner. Referring back to our original labeling of the skewb, the orbits are the odd corners {1, 3, 5, 7}, and the even corners {2, 4, 6, 8}. Let the orbit of odd corners be denoted by V (odd), and let V (even) denote the orbit of even corners. We now partition GV so that GV = GV (odd) × GV (even) . We know that each orbit maps to a permutation on 4 vertices and an orientation on 4 vertices. So GV is a subgroup of (C34 >¢ S4 ) × (C34 >¢ S4 ). Let h = (s, u(h), t, v(h)) be an element of GV . First we will examine the permutations of the vertices of both orbits. Each generator produces a 3-cycle on the vertices, whether they are in the odd or even orbit. Therefore, we can say that the permutations of each orbit generate A4 by the same argument used for the center permutations. Now, GV is a subgroup of (C34 >¢ A4 ) × (C34 >¢ A4 ).

218

CHAPTER 11. OTHER RUBIK-LIKE PUZZLE GROUPS

Claim 1: There exist h, such that s is an element of A4 , u(h) = (0, 0, 0, 0), t is an element of A4 , and v(h) = (0, 0, 0, 0). We know this is true because there are clean skewb moves which only permute 3 vertices. Claim 2: Given any permutation u0 of (1, 2, 0, 0) by an element of A4 and any permutation v 0 of (1, 2, 0, 0) by an element of A4 , there exist h, such that s = 1, u(h) = u0 , t = 1, and v(h) = v 0 . This is true because there are clean skewb moves which only twist vertices. If we combine the moves of Claims 1 and 2, we should generate all of the possible moves of GV . The condition on each of the vertex 4-tuples will drop them in dimension to elements of C33 . So we can conclude that GV is a subgroup of index 9 of (C34 >¢ A4 ) × (C34 >¢ A4 ). Note: This claim is verified by GAP. Since GC = A6 , and GV = (C33 >¢ A4 ) × (C33 >¢ A4 ), we can conclude that G = A6 × (C33 >¢ A4 ) × (C33 >¢ A4 ). and |G| = (6!/2) ∗ (4!/2)2 ∗ (36 ) = 37, 791, 360. In conclusion, it is interesting to note that if we let G0 denote the illegal skewb group - where reassembly is permitted - then G0 = S6 × S8 × C38 . and |G|/|G0 | = .0001984127... This means that if you could take apart the skewb and reassemble it however you wanted, only .02would be solvable. As M. Sch¨onert points out in a post to [CL], this makes the skewb harder to solve that the Rubik’s cube in some sense. Permutation and Orientation Tables

11.2. ON THE GROUP STRUCTURE OF THE PYRAMINX Move UFR UFL DFR DFL BRU BLU DBR DBL

219

Center Permutation Vertex Permutation (1 5 2) (2 6 4) (1 4 5) (1 7 3) (1 2 6) (1 5 7) (1 6 4) (4 6 8) (2 5 3) (1 3 5) (3 5 4) (2 4 8) (2 3 6) (2 8 6) (3 4 6) (3 7 5) Move Vertex Orientation UFR (1 2 0 2 0 2 0 0) UFL (2 0 2 1 0 0 2 0) DFR (2 0 0 0 2 1 2 0) DFL (0 0 0 2 0 2 1 2) BRU (2 1 2 0 2 0 0 0) BLU (0 2 1 2 0 0 0 2) DBR (0 2 0 0 1 2 0 2) DBL (0 0 2 0 2 0 2 1) UFR*UFL (0 2 2 2 0 0 2 0) DFR*DFL (2 0 0 2 0 0 2 2)

Note: The orientations for the generator moves contain two repeated orbits - permutations of (1 0 0 0) and permutations of (2 2 2 0). Also, the combination moves are equivalent to 0 mod 3.

11.2

On the group structure of the pyraminx

The results of this section were worked out with Ann Luers as part of an honors thesis [Lu]. Notation: Let • V denote the vertices of the tetrahedron (which we identify with the set of corner pieces of the pyraminx), • E denote the edges of the tetrahedron (which we identify with the set of edge pieces of the pyraminx), • C the set of interior pieces of the tetrahedon (ie, movable pieces of the pyraminx not in E or V ),

220

CHAPTER 11. OTHER RUBIK-LIKE PUZZLE GROUPS

• SV the permutation group of V , • AV the alternating group of V , • SE the permutation group of E, • AE the alternating group of E. Assume that the tetrahedron is lying on a flat surface in front of you, with the triangle base pointing away from you. The corners are denoted L (left), R (right), U (up), and B (back). Basic Moves: Opposite each corner or vertex there are three layers: the tip, the middle layer, and the opposite face. Let • l denote the 120 degree clockwise rotation of the tip containing the left corner, • L denote the 120 degree clockwise rotation of the tip/middle layer containing the left corner, • r denote the 120 degree clockwise rotation of the tip containing the right corner, • R denote the 120 degree clockwise rotation of the tip/middle layer containing the right corner, • u denote the 120 degree clockwise rotation of the tip containing the up corner, • U denote the 120 degree clockwise rotation of the tip/middle layer containing the up corner, • b denote the 120 degree clockwise rotation of the tip containing the back corner, • B denote the 120 degree clockwise rotation of the tip/middle layer containing the back corner. Let G =< R, L, U, B, r, l, u, b > denote the pyraminx group. Each move g ∈ G induces a permutation of E denoted σ(g). Note that G does not permute the vertices. Furthermore, the tip moves r, l, u, b do not effect the edges.

11.2. ON THE GROUP STRUCTURE OF THE PYRAMINX

221

Lemma 210 σ : G → SE is a group homomorphism. Example 211 ρ(L) is a 3-cycle in SV , σ(L) is a 3-cycle in SE .

11.2.1

Orientations

Assume for the moment that the pyraminx is fixed in space as above and is in the ”solved” position. For each corner or edge piece, choose once and for all one facet on that piece. There are three possible choices for each corner piece and two for the edges. Mark each of these choosen facets with an imaginary ’+’, leaving the other facets unmarked. For the rest of this section, we shall make the following choices for the marked facets (with reference to the numbering in §4.10 ): • marked edge facets: 4, 6, 10, 15, 20, 25 • marked corner facets: 1, 13, 17, 23 For each edge piece, assign to a move g ∈ G either • a ’0’ if the ’+ facet’ for that piece when it was in the solved position is sent to the ’+ facet’ for that piece when it was in the present position, • a ’1’ otherwise, This yields a 6-tuple of 0’s and 1’s: w(g) ~ = (w1 , w2 , ..., w6 ). Example 212 We compute the effect of the basic twist moves on the edge orientations: X B R L U R ∗ U −1 ∗ R−1 ∗ U

w(X) ~ (0,0,0,1,0,1) (0,0,1,0,1,0) (1,0,1,0,0,0) (0,1,0,1,0,0) (1,1,0,0,0,0)

For each corner piece, assign to a move g ∈ G either • a ’0’ if the ’+ facet’ for that piece when it was in the solved position is sent to the ’+ facet’ for that piece in the present position,

222

CHAPTER 11. OTHER RUBIK-LIKE PUZZLE GROUPS

• a ’1’ if the ’+ facet’ for that piece when it was in the solved position is sent to the facet which is a 120 degrees rotation about its vertex from the ’+ facet’ for that piece in the present position, • a ’2’ otherwise, thus yielding a 4-tuple of 0’s, 1’s, and 2’s: ~v (g) = (v1 , v2 , v3 , v4 ). Example 213 We compute the effect of the basic twist moves on the corner orientations: X B R L U

~v (X) (0,0,0,2) (0,0,2,0) (0,2,0,0) (2,0,0,0)

Proposition 214 If w(g) ~ = (w1 , w2 , ..., w6 ) corresponds to a move g ∈ G then w1 + w2 + ... + w6 ≡ 0(mod 2). Observation: There is no corresponding condition for the v1 , ..., v4 , since corner moves move them around freely. proof: The proof uses the following lemma, but is otherwise essentially the same as the corresponding fact (Theorem 202 (c)) which we proved for the Rubik’s cube. The modifications required for the proof are left to the student as an exercise to test their understanding of the argument. 2 Lemma 215 For g, h ∈ G, we have w(g ~ ∗ h) = σ(g)−1 (w(h)) ~ + w(g). ~ proof: The proof is essentially the same as the corresponding fact (Lemma 199) which we proved for the Rubik’s cube. The modifications required for the proof are left to the reader as an exercise. 2 Lemma 216 For g, h ∈ G, we have ~v (g ∗ h) = ~v (h) + ~v (g). Exercise 11.2.1 Prove this lemma.

11.2. ON THE GROUP STRUCTURE OF THE PYRAMINX

223

Let H denote the enlarged pyraminx group generated by G and the ”illegal edge moves” (that is, one may physically remove the edge pieces and reassemble the pyraminx. Illegal center or corner moves are not allowed in H. Let H ∗ = {(s, x, y) | r ∈ SV , s ∈ SE , x ∈ C34 , y ∈ C26 } and define ∗ : H ∗ × H ∗ → H ∗ by (s, x, y) ∗ (s0 , x0 , y 0 ) = (s ∗ s0 , x + x0 , s0 (y) + y 0 ). Theorem 217

• H ∗ is, with this operation, a group.

• There is are isomorphisms H∼ = H∗ ∼ = C34 × (C26 >¢ SE ), and hence between H and the direct product of the tip moves C34 and the wreath product C34 × (SE wr C2 ). • The map G → H ∗ defined by g 7−→ (σ(g), ~v (g), w(g)), ~ is a homomorphism.

11.2.2

Center pieces

Each corner piece has 3 center pieces neighboring it. Facts: • These center pieces, in the middle layer down from the corner and never be moved into any other corner’s middle layer. • The center pieces associated to a corner can never be moved into a middle layer associated to another corner. • The center pieces associated to a corner can always be color-aligned with the colors of the corner piece by a corner twist move. The third part says, in other words, that the center pieces can always be ”solved” by a corner piece.

224

CHAPTER 11. OTHER RUBIK-LIKE PUZZLE GROUPS

11.2.3

The group structure

Theorem 218 G is isomorphic to {(s, x, y) ∈ H ∗ | s even, y1 + y2 + ... + y6 ≡ 0(mod 2)}. The idea to prove this is to show that • AE =< σ(R), σ(L), σ(U ), σ(B) > • G → AE , g 7−→ σ(g), is surjective, • G → {(w1 , ..., w6 ) ∈ C26 | w1 + ... + w6 ≡ 0(mod 2)}, g 7−→ w(g), ~ is surjective (as a map of sets). Here’s the proof of the first point: We can label (as in §5.10 ) the edges 1, 2, ..., 6 so that the edges on the front face are 1, 2, 3, resp., where the f l edge is 1, the f r edge is 2, and the f d edge is 3 (here f , r, l, and d denote the front face, right face, left face, and down face, resp.). The move [R, U −1 ] = R ∗ U −1 ∗ R−1 ∗ U is the counterclockwise cyclic permutation (1, 3, 2) of the edges on the f face. (This move does not affect any corners but does flip some edges, a fact which we may ignore for now since we are only concerned with the permutations now.) In particular, (1, 3, 2) may be written as a product of the generators in {σ(R), σ(L), σ(U ), σ(B)}. Now pick any i ∈ {4, 5, 6} and let s ∈ G denote a move which sends edge i to edge 2 and does not move edge 1 or 3. The move s ∗ [R, U −1 ] ∗ s−1 is the 3-cycle (1, 3, i). It does not affect any corners or other edges. By Lemma 159, these permutations generate A6 ∼ = AE . 2 The second point follows immediately from the first point proven above. Here’s the proof of the last point: The move g = [R, U −1 ] has the following effect on the orientation: w(g) = (1, 1, 0, 0, 0, 0). The group {(w1 , ..., w6 ) ∈ C 6 | w1 + ... + w6 ≡ 0(mod 2)} ∼ = C5 2

2

is a vector space over F2 . The 5 vectors listed in the table for the values for w are all independent. It follows from this and the group law for G proves that the map g 7−→ w(g) ~ is surjective. 2 The theorem 218 above is thus proven.

11.3. A UNIFORM APPROACH

11.3

225

A uniform approach

This section shall follow [GT] in a uniform discussion of the pyraminx (tetrahedron), the 3 × 3 Rubik’s cube, and the megaminx (dodecahedron). Other puzzle groups are analyzed in [GT] (see also [B], chapter 2, [NST], chapter 19). Notation: Let • Gp (resp., GR , Gm ) denote the permutation puzzle group generated by the basic moves of the pyraminx (resp., the Rubik’s cube, megaminx), • Vp (resp., VR , Vm ) denote the set of vertex pieces of the pyraminx (resp., the Rubik’s cube, megaminx), • Ep (resp., ER , Em ) denote the set of edge pieces of the pyraminx (resp., the Rubik’s cube, megaminx), • Fp (resp., FR , Fm ) denote the set of facets of the movable pieces of the pyraminx (resp., the Rubik’s cube, megaminx).

11.3.1

General remarks

Let G, V, E, F (resp.) denote either Gp , Ep , Vp , Fp (resp.), or GR , ER , VR , FR (resp.), or Gm , Em , Vm , Fm (resp.). Lemma 219 G acts on the set V , (resp., E, F ). If g is any move in G then, since g acts on the sets V , E, and F , we may regard g • as an element of the symmetric group SV of V , • as an element of the symmetric group SE of E, or • as an element of the symmetric group SF of F . These groups SV , SE , and SF are different, so to distinguish these three ways of regarding g, let us write • gV for the element of SV corresponding to g, • gE for the element of SE corresponding to g,

226

CHAPTER 11. OTHER RUBIK-LIKE PUZZLE GROUPS

• gF for the element of SF corresponding to g. What is the kernel of fV ? What is its image? To answer this question (actually, we shall not answer this precise question but one similar to it) we introduce a certain subgroup of the symmetric group. Let SX denote the symmetric group of a finite set X. Recall the alternating group AX is the subgroup of all permutations of X which are even as an element of SX (in the sense of example 81 above).

11.3.2

Parity conditions

Consider the function

fV E : G → SV × SE g 7−→ (gV , gE )

It is easy to check that this is a homomorphism. Theorem 220 The image fV E (G) of fV E is isomorphic to (

AV × AE , for the pyraminx, megaminx, {(x, y) ∈ SV × SE | x, y both even or both odd}, for Rubik0 s cube.

This is a consequence of a result proven below in the case of the Rubik’s cube. To see what this theorem means, we look at an example. Example 221 Let G = GR . Question: Can you find a move of the Rubik’s cube which flips a single edge subcube over, leaving the rest of the puzzle pieces unmoved? If so, then the image of fEV would have to contain an element (x, y) with x = 1 (since moving an edge only does not effect the vertices) and where y is a 2-cycle. But x = 1 is even and a 2-cycle is odd. This contradicts the theorem, which says that x, y are either both even or both odd. Therefore, the answer is no: a single edge flip is impossible. Exercise 11.3.1 Is there a move on the Rubik’s cube which flips two corner subcubes over, without rotation (so it is a 2-cycle on the edges connected to the corner vertices being swapped), leaving the rest of the puzzle pieces unmoved?

11.4. THE HOMOLOGY GROUP OF THE SQUARE 1 PUZZLE

227

Next, some more notation: let K = ker(fV E ) < G denote the kernel of the map fV E introduced above. This is a normal subgroup of G. Example 222 In the case of the Rubik’s cube, this subgroup K is the set of moves which may reorient (i.e., flip or rotate) a subcube but does not swap it with some other subcube. For example, the move ((D2 )R ∗ (U 2 )B )2 , which twists the ufr corner clockwise and the bld corner counterclockwise, belongs to K. (Here xy = y −1 ∗ x ∗ y, x2 = x ∗ x.) Theorem 223 (Gold, Turner [GT]) G is a semi-direct product of K with fV E (G). This is a consequence of a result proven below in the case of the Rubik’s cube. In the case of the 3 × 3 Rubik’s cube, some more details are given in the next few sections. See also [GT] or [NST], chapter 19.

11.4

The homology group of the square 1 puzzle

This section is based on a paper written jointly with J. McShea [JM]. Here we study the group theoretic properties of the collection G of all ”words” in the basic moves of the square 1 puzzle which preserve the cube shape. This collection G forms a group which, motivated by [W], we call the homology group of the square 1 puzzle. The list of shapes which the square 1 puzzle can make is given in [Sn2]. It is not hard to see that the homology group of any one of these other shapes is conjugate to G, so from a group-theoretic standpoint, we may focus our attention on the cube. We shall also make use of the moves given in [Are] which belong to G.

228

CHAPTER 11. OTHER RUBIK-LIKE PUZZLE GROUPS

A bi-product of the proof is an collection of moves which can be used to solve the square 1 puzzle, once it is put in the cube shape.

11.4.1

The main result

Let Sn denote the symmetric group of degree n, i.e., the group of permutations of {1, 2, ..., n}. Let sgn : Sn → {±1} denote the homomorphism which assigns to each permutation its sign (the sign of a cyclic permutation of length r is (−1)r+1 , for example). We shall see that the size of the homology group of the square 1 is about .8 billion. Theorem 224 G is isomorphic to the kernel of index 2 in S8 × S8 of the homomorphism f : S8 × S8 → {±1} defined by f (g1 , g2 ) = sgn(g1 )sgn(g2 ). Consequently, |G| = 213 34 52 72 = 812851200. As a corollary of the proof of this theorem, given below, we shall see that any even permutation of the corners is possible and any even permutation of the wedges is possible.

11.4. THE HOMOLOGY GROUP OF THE SQUARE 1 PUZZLE

229

Let H denote the enlarged square 1 group generated by all legal moves preserving the cube shape and all illegal moves (i.e., disassembly and reassembly is allowed) preserving the cube shape. It is clear that H∼ = S8 × S8 .

Some notation We shall assume that the puzzle is in the solved position with the ”square 1” side in front, right-side up. Let

• u denote rotation of the up face by 30o clockwise,

• d denote rotation of the up face by 30o clockwise,

• R denote rotation of the cube by 180o though one of the skew-diagonal cuts (in a given position, at most one such move is possible, so this is unambiguous).

. Like the 15 puzzle, and unlike the Rubik’s cube, not any sequence of u, d, and R’s is possible. Let T (x, y) = u ∗ R ∗ x ∗ y ∗ R ∗ u−1 , B(x, y) = d−1 ∗ R ∗ x ∗ y ∗ R ∗ d, where x, y are moves of the square 1 puzzle.

230

CHAPTER 11. OTHER RUBIK-LIKE PUZZLE GROUPS

In the notation of these diagrams, we have uRu−1 d−1 Rd = (2, 8)(4, 6) T (u3 , 1) = (10 , 60 , 70 , 40 )(1, 6, 7, 4) T (1, d3 ) = (20 , 30 , 80 , 50 )(2, 3, 8, 5) B(u3 , 1) = (1, 2, 7, 8)(10 , 60 , 70 , 40 ) B(1, d3 ) = (3, 4, 5, 6)(20 , 30 , 80 , 50 ). Two subgroups Let Gu =< T (u3 , 1), T (1, d3 ) > and Gd =< B(u3 , 1), B(1, d3 ) > Lemma 225 Gu and Gd are each isomorphic to C4 × C4 . proof: We have T (u3 , 1)T (1, d3 ) = T (1, d3 )T (u3 , 1). Moreover, T (u3 , 1) and T (1, d3 ) are each of order 4. Since C4 × C4 =< a, b | a4 = 1, b4 = 1, ab = ba >,

11.4. THE HOMOLOGY GROUP OF THE SQUARE 1 PUZZLE

231

the lemma follows. 2 The homology group of the square 1 puzzle is defined to be G =< d3 , u3 , B(u3 , 1), B(1, d3 ), T (u3 , 1), T (1, d3 ) > We shall use the following labelings to describe the moves of the square 1 puzzle

11.4.2

Proof of the theorem

We shall prove the theorem in the following steps: • Show that the wedge 3-cycle (1, 2, 3) and the corner 3-cycle (10 , 20 , 30 ) each belong to G. • Show that any wedge 3-cycle (1, 2, i) and each corner 3-cycle (10 , 20 , i0 ) belong to G. • Show that there is a injective homomorphism φ : G → S8 × S8 where the image φ(G) contains A8 × A8 . • Conclude that G ∼ = S8 × S8 /{±1}. Step 1: First, we claim that (1, 2, 3) belongs to G. In fact, the 3-cycle (1, 2, 3) is obtained from the move M1 = (B(u3 , 1)∗d3 )∗((B(u3 , 1)∗d−3 )∗(B(u−3 , 1)∗T (1, d−3 )∗d6 )))4 ∗(B(u3 , 1)∗d3 )−1 . (Incidently, this 80 move long manuever may be verified using GAP [Gap]. See also [Sn2].) Next, we claim that (10 , 20 , 30 ) belongs to G. In fact, M2 = Ru3 Rd−3 Ru3 (Ru−3 )2 d3 Ru−3 is the product of 2-cycles (20 , 30 )(3, 4). (This move was found in [Sn2].) Therefore, u3 M2 u−3 is the product of 2-cycles (10 , 20 )(2, 3). The product of these is (10 , 20 , 30 )(2, 3, 4). Since (2, 3, 4) is obtained from u−3 M1 u3 , we see that (10 , 20 , 30 ) is in G. (This may also be verified using GAP.) Step 2: Let g be any move in G which sends wedges 3 to wedge i, resp., and does not move wedges 1, 2 (it may permute other wedges and corners). Then (1, 2, i) = g ∗ (1, 2, 3) ∗ g −1 . Thus (1, 2, i) ∈ G.

232

CHAPTER 11. OTHER RUBIK-LIKE PUZZLE GROUPS

The proof that each (10 , 20 , i0 ) ∈ G is similar. Step 3: It is clear from our definition that there is an injection G → S8 ×S8 as sets. The verification that this is a homomorphism is straightforward. Step 4: The group A8 is generated by the 3-cycles (1, 2, i) (see Lemma 159). Since these all belong to G, all even wedge permutations are possible. Similarly, all even corner permutations are possible. Thus A8 × A8 ⊂ G. Let p1 : S8 × S8 → S8 denote the projection onto the first factor. Let p2 denote the projection onto the second factor. For each generator g ∈ {d3 , u3 , B(u3 , 1), B(1, d3 ), T (u3 , 1), T (1, d3 )} of G we have sgn(p1 (g)) = sgn(p2 (g)). Thus the image φ(G) is strictly contained in S8 × S8 . In fact, this shows that φ(G) is contained in the kernel ker(f ) of the homomorphism f : S8 × S8 → {±1} defined in the statement of the theorem. Since A8 × A8 ⊂ G ⊂ ker(f ), [ker(f ) : A8 × A8 ] = 2, and T (u3 , 1) ∈ / A8 × A8 , the theorem follows. 2

Chapter 12 Interesting subgroups of the cube group ”[Lefschetz and Einstein] had a running debate for many years. Lefschetz insisted that there was difficult mathematics. Einstein said that there was no difficult mathematics, only stupid mathematicians. I think that the history of mathematics is on the side of Einstein.” Richard Bellman EYE OF THE HURRICANE, 1984

It is remarkable that several ”familiar” groups may be embedding into the Rubik’s cube group, and hence be regarded as a subgroup of the cube group. For example, we have seen in an earlier chapter how to embed the group of quaternions Q = {1, −1, i, −i, j, −j, k, −k} inside the Rubik’s cube group. The subgroup method, discussed in the appendix, is a method for investigating God’s algorithm using a computer. One of the groups arising in this method is the group studied in the first section of this chapter. The second section studies the ”two faces” group. 233

234CHAPTER 12. INTERESTING SUBGROUPS OF THE CUBE GROUP

12.1

The squares subgroup

Let G denote the subgroup of the Rubik’s cube group generated by the squares of the basic moves: G :=< U 2 , D2 , R2 , L2 , F 2 , B 2 > called the squares group. We shall verify that the order of this group is 213 34 . (As a consequence of Burnside’s theorem [R], ch. 5, it follows that G is a solvable group.) In this section, we will investigate the group struction of G using the same method which was used to determine the structure of the Rubik’s cube group. The group G acts on the set of edges and the set of vertices of the cube. There is a choice of orientation of the edges (resp., corners) similar to that in §§10.1.2-3 such that each element of G preserves the edge (resp., corner) orientations. The action of G on the edges E of the cube has exactly 3 orbits: the middle slice parallel to the right face ER , the middle slice parallel to the front face EF , the middle slice parallel to the up face EU . The action of G on the set of vertices V has exactly 2 orbits: V1 = {uf r, ubl, df l, drb}, V2 = {uf l, ubr, df r, dlb}. We have homomorphisms: φ : G → SE , φER : G → SER , φEF : G → SEF , φEU : G → SEU , ψ : G → SV , ψ1 : G → SV1 , ψ2 : G → SV2 . Proposition 226 G = φ(G) × ψ(G). Lemma 227 If g ∈ G then sgn(φER (g))sgn(φEF (g))sgn(φEU (g)) = 1. Conversely, if (p1 , p2 , p3 ) ∈ SER × SEF × SEU then there is a g ∈ G such that p1 = φER (g), p2 = φEF (g), p3 = φEU (g) if and only if sgn(p1 )sgn(p2 )sgn(p3 ) = 1.

12.1. THE SQUARES SUBGROUP

235

As a consequence, we find that φ(G) = ker(sgn × sgn × sgn : φER (G) × φEF (G) × φEU (G) → {±1}) ∼ = (S4 × S4 × S4 )/C2 . In particular, |φ(G)| = (4!)3 /2 = 28 33 . It remains to determine ψ(G). We denote this group by H for notational simplicity. We may label the vertices of the cube 1, 2, ..., 8 in such a way that H =< u, d, l, r, f, b >, where u = (1, 3)(2, 4), f = (1, 8)(4, 5), d = (5, 7)(6, 8), b = (3, 6)(2, 7), r = (2, 5)(1, 6), l = (4, 7)(3, 8). The action of H on the set of vertices of the cube has two orbits ({1, 3, 6, 8} and {2, 4, 5, 7} in our labeling above), which we denote for simplicity by V1 and V2 . There are homomorphisms ψ1 : H → SV1 ,

ψ2 : H → SV2 ,

but we shall not say much about these. Instead, we use GAP [Gap] to determine more about H. According to GAP, this group has |H| = 96 elements and 10 conjugacy classes (by the way, GAP also says that all the generators u, ..., b are conjugate): size representative 1 1 12 d=(5,7)(6,8) 32 l*d=(3,6,8)(4,5,7) 3 d*l*b*l=(2,4)(5,7) 12 d*b*l=(2,4,7,5)(3,6) 3 l*b*l*u=(1,3)(6,8) 12 b*l*u=(1,3,6,8)(4,7) 3 u*d=(1,3)(2,4)(5,7)(6,8) 12 l*d*u=(1,3,6,8)(2,4,5,7) 6 u*r*d*f=(1,3)(2,5)(4,7)(6,8) The stabilizer in H of any vertex v ∈ V , written Hv , is a subgroup of order 24 isomorphic to the symmetric group S4 . Furthermore, H has a normal subgroup N of order 48 (and index 2), where N is a semidirect product of C3 by C24 , with C24 normal in N . This is all we shall say about H. The order of G is therefore G = |φ(G)| · |H| = 96 · (4!)3 /2 = 213 34 , as claimed above.

236CHAPTER 12. INTERESTING SUBGROUPS OF THE CUBE GROUP

12.2

P GL(2, F5) and two faces of the cube

The material in this section was communicated to me by Dan Bump (the idea originally arose in [Si], I believe). This section is relatively advanced in that it requires more mathematical background from the reader than the previous chapters. This section is devoted to ”determining” the two-face group generated by only two basic moves, < F, U >. D. Singmaster [Si] has shown that < F, U >∼ = S7 × P GL2 (F5 ), where P GL2 (F5 ) is a group of order 120 which is defined below. Here S7 arises from the action of the Rubik’s cube group on the edges and P GL2 (F5 ) arises from the action on the corners. In this chapter, we focus on the action on the corners.

12.2.1

Finite fields

In this subsection, we introduce fields and especially finite fields. The general definition A field is a set F with an addition law + and a multiplication law · which obeys the a list of properties similar to those for the field of real numbers R. More precisely, we call (F, +, ·) a field if (F1) (F, +) is an abelian group, with an identity element denoted 0 (”the additive group of the field”), (F2) for all x, y, z ∈ F , (x + y)z = xz + yz (”distributive law”), (F3) (F − {0}, ·) is an abelian group, with an identity element denoted 1 (”the multiplicative group of the field”). It happens to be true that if F is a finite field then not only is (F − {0}, ·) an abelian group, it is actually a cyclic group. Definition 228 Let F1 , F2 be wo fields. A function f : F1 → F2 is called a field isomorphism if (a) when f is restricted to the additive group (F1 , +), call this restriction f again, it yields an isomorphism of groups f : (F1 , +) → (F2 , +),

12.2. P GL(2, F5 ) AND TWO FACES OF THE CUBE

237

(b) when f is restricted to the multiplicative group (F1 − {0}, ·), call this restriction f again, it yields an isomorphism of groups f : (F1 − {0}, ·) → (F2 − {0}, ·). A construction of Fp Recall from §2.3 that congruence modulo n (n > 1 an integer) is an equivalence relation. Let n = p be a prime and let k denote the equivalence class of k with respect to this equivalence relation. Let Fp denote the finite field with p elements, so Fp is, as a set, Fp = {0, 1, ..., p − 1}, with addition and multiplication being performed mod p. Example 229 When p = 5, F5 will denote the finite field with 5 elements, so F5 is, as a set, F5 = {0, 1, 2, 3, 4}, with addition and multiplication being performed mod 5. It is a general fact that if F is any finite field then there is a prime number p such that px = 0 for all x ∈ F . This prime number is called the characteristic of F . The easiest example of a finite field with characteristic p is the finite field having p elements, Fp . It is not hard to see that any other finite field F of characteristic p must be a finite dimensional vector space over Fp . (Even if you’ve never seen a ”vector space over Fp ” defined before, if you know what a ”real vector space” is then you’ve got the right idea.) The dimension of the vector space F is called the degree of F over Fp , denoted d = [F : Fp ], and F is called a field extension of Fp of degree d. It is a general fact that for fixed p, d there is, up to isomorphism, only one such field. Next, we shall show how to construct in a very simple way such extensions. A construction of finite fields First, some general remarks. Since F is a finite dimensional vector space containing Fp , it has a vector space basis which we label as e1 = 1, e2 , ..., ed .

238CHAPTER 12. INTERESTING SUBGROUPS OF THE CUBE GROUP Thus F is, as a set, the collection of elements of the form x1 e1 + ...xn en ,

xi ∈ Fp .

Since F is a field, there are ckij ∈ Fp , which we call structure constants, such that n ei ej =

X

ckij ek .

k=1

There are dki ∈ Fp , which we call inversion constants, such that e−1 i =

n X

cki ek .

k=1

These constants determine how to multiply and divide elements of F . We shall consider F ”completely described” once we explicitly determine these constants. Example 230 Let p = 5, so F5 = {0, 1, 2, 3, 4}. The set of squares is given by {x2 | x ∈ F5 } = {0, 1, 4}. √ In particular, 2, 3 are not squares in this field. Let e2 = 2 be a formal symbol for some element which satisfies e22 = 2. This is a root of the polynomial x2 − 2 = 0. The vector space F over F5 with basis √ {e1 = 1, e2 } is 2-dimensional √ over F5 . Two elements x1 e1 + x2 e2 = x1 + x2 2 and y1 e1 + y2 e2 = y1 + y2 2 are multiplied by the rule √ √ √ (x1 + x2 2) · (y1 + y2 2) = x1 y1 + 2x2 y2 + (x1 y2 + y1 x2 ) 2. It is a degree 2 field extension since c111 = 1, c211 = 0, c112 = 0, c212 = 1, c121 = 2, c221 = 0, and d11 = 1, d21 = 0, d12 = 0, d22 = 3. The construction used in the above example may be summarized more generally as follows:

12.2. P GL(2, F5 ) AND TWO FACES OF THE CUBE

239

1. Pick an element m ∈ Fp which is not the square of another element. 2. Let e1 = 1 and e2 = satisfies e22 = m.



m be a formal symbol for some element which

3. As a set, let F = {xe1 + ye2 | x, y ∈ Fp }. To define F as a field, let + be ”componentwise addition” mod p and let · be defined by √ √ √ (x1 + x2 m) · (y1 + y2 m) = x1 y1 + mx2 y2 + (x1 y2 + y1 x2 ) m. A finite field F constructed in this way is called a quadratic extension of Fp . More generally, let d > 1 be an integer. 1. Pick an element m ∈ Fp which is not the dth power of another element. 2. Let e1 = 1, let e2 = m1/d be a formal symbol for some element which satisfies ed2 = m, and (if d > 2) let ei = ei−2 for i = 3, ..., d. 2 3. As a set, let F = {x1 e1 + ... + xd ed | xi ∈ Fp }. To define F as a field, let + be ”componentwise addition” mod p and let · be defined by expanding and collecting (x1 e1 + ... + xd ed ) · (y1 e1 + ... + yd ed ). A finite field F constructed in this way is called a degree d extension of Fp . It has pd elements. It turns out that any two fields having pd elements must be isomorphic. Therefore, any finite field must be isomorphic to one described above. Definition 231 The projective plane P1 (Fp ) = {0, 1, ..., p − 1, ∞} is defined to be the set of lines through the origin in the Cartesian plane F2p , associating each number (including ∞) with the slope of the corresponding line.

240CHAPTER 12. INTERESTING SUBGROUPS OF THE CUBE GROUP

12.2.2

M¨ obius transformations

If a, b, c, d ∈ Fp are given numbers (not all equal to zero) then we define the Mobius transformation f by: f : P1 (Fp ) → P1 (Fp ) . x 7−→ ax+b cx+d Ã

Theorem 232 f is a bijection if and only if det

a b c d

!

6= 0.

Before proving this, we need the following Definition 233 Define Ã

GL(2, F5 ) = {

a b c d

!

| a, b, c, d ∈ F5 , ad − bc 6= 0}.

This set is a group under ordinary matrix multiplication and, furthermore, acts on the set P1 (F5 ) by means of Mobius transformations thus defining a function à ! a b : P1 (F5 ) → P1 (F5 ) c d Lemma 234 (a) The center of GL(2, Fp ) (i.e., the subgroup of all elements which commute with every element in GL(2, Fp )) is given by Ã

Z(GL(2, Fp )) = {

a 0 0 a

!

| a ∈ Fp , a 6= 0}.

(b) This subgroup is normal in GL(2, Fp ). (c) There is an isomorphism Z(GL(2, Fp )) ∼ = F× p. proof: (a) Since Ã

r 0 0 r



a b c d

!

Ã

=

r 0 0 r



a b c d

!

12.2. P GL(2, F5 ) AND TWO FACES OF THE CUBE

241

we can conclude that Ã

a 0 0 a

{

!

| a ∈ Fp , a 6= 0} ⊂ Z(GL(2, Fp )).

To show that Ã

Z(GL(2, Fp )) ⊂ { assume that

Ã

r s u v



a b c d

a 0 0 a !

!

Ã

=

| a ∈ Fp , a 6= 0},

r s u v



a b c d

!

for all a, b, c, d. This implies bu = cs for all b, c. This is impossible unless u = s = 0. This in turn forces cr = cv, for all c. This implies r = v. This proves the desired inclusion. The proof of parts (b) and (c) are left as an exercise for the reader. 2 Definition 235 The quotient group, denoted P GL(2, Fp ) = GL(2, Fp )/Z(GL(2, Fp )), is called the projective linear group. (This is a group since the center is a normal subgroup by the lemma above.) Lemma 236 This group P GL(2, Fp ) acts on the set P1 (Fp ) by means of the linear fractional transformations. Remark 23 In fact, the action of P GL(2, Fp ) on the set P1 (Fp ) is 3-transitive. This not hard to prove but we left it to the interested reader to look it up in [R] (see Theorem 9.48). proof: First, we show that the GL(2, Fp ) acts on the set P1 (Fp ) by means of the linear fractional transformations. In other words, if Ã

φ then

Ã

(a) φ Ã

φ

1 0 0 1

!

1 0 0 1

a b c d

!

(x) =

ax + b cx + d

!

(x) = x, for all x (i.e., the linear fractional transformation

is the identity map),

242CHAPTER 12. INTERESTING SUBGROUPS OF THE CUBE GROUP (b) φ(A) ◦ φ(B) = φ(AB), for all A, B ∈ GL(2, Fp ). We leave (a) to the reader and check (b). Let A = Ã

r s u v

!

Ã

a b c d

!

and B =

. Then Ã

AB =

ar + bu as + bv cr + du cs + dv

!

,

so φ(AB)(x) =

(ar + bu)x + as + bv . (cr + du)x + cs + dv

On the other hand, φ(A)(φ(B)(x)) is equal to φ(A)(

rx+s )+b a( ux+v rx + s ) = rx+s . ux + v c( ux+v ) + d

Simplifying this, we see that the last two displayed equations are equal. This verifies (b). Therefore, GL(2, Fp ) acts on the set P1 (Fp ). Ã ! Ã ! a b ra rb Let Z = Z(GL(2, Fp )). Since φ = φ , for all nonc d rc rd zero r, it follows that we may define an action of P GL(2, Fp ) on the set P1 (Fp ) by φ(A · Z) = φ(A), for all A ∈ GL(2, Fp ). 2 proof of the theorem: Let φ be as above and let f be as in the statement of the theorem. (⇐): Since GL(2, Fp ) acts on the set P1 (Fp ), we have 1 = φ(AA−1 ) = φ(A)φ(A−1 ), so φ(A) is invertible. This implies that f isÃa bijection. ! a b = 0. By a (⇒): We prove the contrapositive. Suppose that det c d result in linear algebra (see any text book, for example [JN]), the row vectors of this matrix are linearly dependent. This implies that there is an r ∈ Fp such that either (a, b) = r · (c, d) or (c, d) = r · (a, b). In either case, the quotient f (x) = ax+b is a constant independent of x, so cannot be surjective. cx+d This proves that f is not a bijection, which verifies the contrapositive. 2

12.2. P GL(2, F5 ) AND TWO FACES OF THE CUBE

12.2.3

243

The main isomorphism

Let G denote the Rubik’s cube group. Let H be the subgroup generated by F and U : H =< F, U > . Exercise 12.2.1 Show the group H acts on the set of vertices above (via the Rubik’s cube group). We describe how to label the six vertices on the ”up” and ”front” faces of the cube, f ru, f lu, df l, df r, bru, blu, with the elements in the projective plane P1 (F5 ) = {0, 1, 2, 3, 4, ∞} in a certain way. More precisely, we will show label the 6 vertices above with elements of P1 (F5 ) in such a way that (a), (b) of the following theorem hold true. Theorem 237 There are a0 , a1 , b0 , b1 , c0 , c1 , d0 , d1 ∈ F5 (given explicitly below) such that (a) the action of F (the usual rotation of the front face) on these vertices is the same as the action of some linear fractional transformation fF (x) =

a 0 x + b0 c0 x + d0

(b) the action of U (the usual rotation of the up face) on these vertices is the same as the action of some linear fractional transformation fU (x) =

a1 x + b1 c1 x + d1

In other words, the basic moves F , U may be regarded as linear fractions transformations over a finite field! Theorem 238 P GL(2, F5 ) =< fF , fU >. Remark 24 P GL(2, F5 ) is isomorphic to S5 (this is part of Exercise 9.25 in [R]). We shall prove these below.

244CHAPTER 12. INTERESTING SUBGROUPS OF THE CUBE GROUP

12.2.4

The labeling

Label the up and front vertices as

Let fF (x) =

x−1 , x+1

fU (x) = 3x + 3.

The map φ : F 7−→ fF ,

U 7−→ FU ,

extends to a surjective homomorphism of groups φ : < F, U > → < fF , fU > ⊂ P GL(2, F5 ).

Exercise 12.2.2 Verify the first theorem above.

12.2. P GL(2, F5 ) AND TWO FACES OF THE CUBE

12.2.5

245

Proof of the second theorem

Let

Ã

denote the image of

a b c d

Ã

!

∈ P GL(2, FF ) ∗

a b c d

!

∈ GL(2, FF )

under the natural map GL(2, F5 ) → P GL(2, F5 ), g 7−→ F× 5 ∗ g. Since à ! 0 −1 2 fU = 1 0 ∗ we have

Ã

0 −1 1 0

!

∈< fU , fF > . ∗

Since

Ã

fF ∗ fU5 = it follows that

Ã

1 0 1 1

−1 0 −1 −1

! ∗

!

∈< fF , fU > . ∗

Conjugating this matrix by fU2 , we find that Ã

1 −1 0 1

!

∈< fF , fU > . ∗

It is known that SL(2, F5 ) is generated by elementary transvections (see [R]). Therefore, P SL(2, F5 ) ⊂< fF , fU >⊂ P GL(2, F5 ). It is also known (see [R]) that |P SL(2, F5 )| = 60 and |P GL(2, F5 )| = 120. It remains to show that there is an element of < fF , fU > which does not belong to P SL(2, F5 ). We claim that such an element is fU . Note that det(fU ) belongs to the set 2 2 × 3(F× 5 ) = {3x | x ∈ F5 }.

246CHAPTER 12. INTERESTING SUBGROUPS OF THE CUBE GROUP But an element of P SL(2, F5 ) must have determinant 1. Since 3−1 = 2 (mod 5) is not a square mod 5, there is no element of F5 which satisfies 1 = 3x2 . Thus fU does not belong to P SL(2, F5 ). 2 Exercise 12.2.3 As an application of theorem 238, use the example in section 5.5 to show that there exists an embedding D12 ,→ P GL(2, F5 ) of the symmetry group of the hexagon into P GL(2, F5 ).

12.3

The cross groups

All the enties in the following table are, as far as I am aware, are new except for the M12 entry. Rubik polyhedra edge cross group vertex cross group tetrahedron A5 ∼ C2 × C2 = P SL2 (F5 ) cube A12 P SL2 (F7 ) octahedron A12 A5 = P SL2 (F7 ) dodecahedron A30 A20 icosahedron A30 A12 rubicon A30 M12 In fact, the subgroup of the dodecahedral edge cross group generated by a subset of the cross moves can yield (smaller but still simple) alternating groups. Problem: (F. Dyson) Work out the analogous cross groups of the ”Rubicized” 4-dimensional regular polyhedra. Problem: Work out the analogous cross groups of the ”Rubicized” 3dimensional Archimedian polyhedra.

12.3.1

P SL(2, F7 ) and crossing the cube

Define a cross move of the cube to be a move of the form X ∗ Y −1 , where X, Y ∈ {R, L, U, D, F, B}. The subgroup generated by the cross moves will be called the cross group. The cross moves permute the set V of vertices of the cube and therefore generate a subgroup C of SV . Theorem 239 C ∼ = P SL2 (F7 ).

12.3. THE CROSS GROUPS

247

The first proof of this is by computer! first proof: Gap [Gap] gives that C is a simple group of order 168. By the classification of simple groups (or, more simply, Exercise 9.26 in [R]), C must be isomorphic to P SL2 (F7 ). 2 The second proof is from [CD]. second proof: P SL2 (F7 ) can be generated by the three matrices: Ã

f1 =

0 −1 1 0

!

Ã

, f2 =

2 1 0 1

!

Ã

, f3 =

2 0 0 1

!

.

We will label the vertices of the cube in the following manner:

labeling the cube by the projective line P 1 (F7 ) Under this labeling, we can show that • the image of the move m1 = (U D−1 )2 will permute the vertices in this way: (∞, 0)(1, 6)(2, 3)(4, 5), The same permutation is given by the mobius transform (0x − 1)/(x + 0) acting on P 1 (F7 ).

248CHAPTER 12. INTERESTING SUBGROUPS OF THE CUBE GROUP • m2 = U R−1 gives us the permutation: (0, 1, 3)(2, 5, 4), which is given by (2x + 1)/(0x + 1) acting of P 1 (F7 ). • m3 = BU −1 LB −1 gives us the permutation (1, 2, 4)(3, 6, 5), which is given by (2x + 0)/(0x + 1) acting on P 1 (F7 ). You should notice that if the constants in these M¨obius transformations (a, b, c, d) are written in matrix form, they correspond to the generators of P SL2 (F7 ). Now we will define a homomorphism q : C → P SL2 (F7 ), such that q(m1 ) = f1 , q(m2 ) = f2 , q(m3 ) = f3 . We want to show that our q is an isomorphism. To do this we will first show that it is surjective. Let f be a matrix in P SL2 (F7 ), which can be written as a product of generators f1 , f2 , f3 (where q(m1 ) = f1 , q(m2 ) = f2 , q(m3 ) = f3 ). Now take f as some element of P SL2 (F7 ). f can be broken down as a product of its generators, f1 , f2 , f3 , we’ll say f=

n Y

fiekk .

k=1

Since we have a homomorphism, we can write it as a product of the images of Q the generators of C. Again we can rewrite it as f = q( nk=1 meikk ). Therefore q is surjective. To show that q is one to one we need to know that P SL2 (F7 ) has order 168 [R], and that the order of the cross group is also 168. (This fact was proven by computer.) We will prove by contradiction that q is one to one. Now we assume that c1 and c2 are elements of C, such that q(c1 ) = q(c2 ), and c1 is not equal to c2 . |P SL2 (F7 )| = |q(C)|. We now subtract c2 from C , and |q(C)| = |q(C − c2 )| because q(c1 ) = q(c2 ). Now we can say that |q(C − c2 )| < or = |C − c2 | because we know that q is surjective. Since we have taken c2 out of C, we know |C − c2 | < |C|, which by transitivity implies |P SL2 (F7 )| < |C|. This is a contradiction because we know |P SL2 (F7 )| = |C|. Therefore q is injective. Now that we have shown that q is both surjective and injective, it is bijective and an isomorphism. 2 The above proof of the theorem tells us explicitly that there exists a labeling of the vertices V of the cube by the elements of the projective line P1 (F) = {∞, 0, 1, 2, 3, 4, 5, 6},

12.3. THE CROSS GROUPS

249

with the property that there is a move c : V → V in C if and only if there is a M¨obius transformation f : P1 (F) → P1 (F) in P SL2 (F7 ). Because the group of M¨obius transformations in P SL2 (F7 ) acts 2-transitively on the projective line P1 (F) (see [R], Theorem 9.45), it follows that we have the following Corollary 240 C acts 2-transitively on V . In other words, for any ordered pairs (v1 , v2 ), (v10 , v2 ) of distinct vertices there is an element c ∈ C sending vi to vi0 , for i = 1, 2.

12.3.2

Klein’s 4-group and crossing the pyramnix

We leave the main result of this section as an Exercise - actually more of a project - for the reader. Exercise 12.3.1 (hard) Show that the subgroup of SV generated by the twistuntwist moves of the pyraminx is isomorphic to the Klein 4 group C2 × C2 . The determination of the cross group for the megaminx is due to J. Conway. It will be presented in the next chapter.

250CHAPTER 12. INTERESTING SUBGROUPS OF THE CUBE GROUP

Chapter 13 Crossing the Rubicon ”Mathematical structures are among the most beautiful discoveries by the human mind. The best of these discoveries have tremendous metaphorical and explanatory power.” Douglas Hofstadter METAMATHEMATICAL THEMAS, 1985

Much of the material here can be found in [CS] and is due to John Conway. The title of this chapter is, however, ”borrowed” from a similarly worded title of an article by D. Hofstadter [H]. More details on parts of this chapter may be found in Ann Luers’ paper [Lu]. This chapter shall be a little more advanced than some of the others. The reader will be assumed to be familiar with some topics covered in a course in linear algebra and elementary number theory or coding theory. We shall also assume some results from Rotman [R], though for the understanding of the material in this chapter the reader may simply take them on faith. Let g1 , ..., g12 denote the basic moves of the Rubik isocahedron. A surprising result of Conway states that the group generated by gi ∗ gj−1 is the simple ”sporadic” group M12 . (This is stated more precisely below.) We shall describe, in this chapter, what M12 is and some of its remarkable properties. They form a basis for my opinion, which I hope you will agree with, that M12 is one of the most interesting objects in mathematics. Let p be a prime unless otherwise stated and let q be a power of p. We shall assume that p > 3. 251

252

13.1

CHAPTER 13. CROSSING THE RUBICON

Doing the Mongean shuffle

Consider a deck of 12 cards labeled 0, 1, ..., 11. Let r, s be the permutations r(t) = 11 − t,

s(t) = min(2t, 23 − 2t).

The permutation r reverses the cards around and the permutation s is called the ”Mongean shuffle”. To perform the reverse shuffle, simply take a stack of cards (face down, say) in your left hand and put them in your right hand one-at-a-time (face down). To perform the Mongean shuffle, take the same stack of cards and, one-at-a-time, put them alternately into one of two piles: the first card face up into the first pile, the second card face down into the second pile, the third card face up into the first pile, the fourth card face down into the second pile, and so on until the pile is exhausted. Now pick up the first pile of face up cards, flip the entire pile over so that they are all fade down and put it on top of the second pile. cards 0 1 2 3 4 5 6 7 8 9 10 11

reverse shuffle 11 10 9 8 7 6 5 4 3 2 1 0

Mongean shuffle 0 2 4 6 8 10 11 9 7 5 3 1

Definition 241 The Mathieu group M12 is defined to be the permutation group M12 =< r, s > < S12 .

13.2

Background on P SL2

We need a few basic facts about the projective special linear group of degree 2. We have already discussed the related group GL(2, Fp ) in the previous chapter, so we refer to there for more details.

13.2. BACKGROUND ON P SL2

253

Definition 242 (1st version) Define SL2 (Fq ) to be the group of all 2 × 2 matrices having entries taken from the finite field Fq and having determinant one. This is called the special linear group of degree 2 over Fq . The center of this group, denoted Z, is the subgroup of 2 × 2 ”scalar” matrices of the form diag(z, z), where z ∈ {1, −1}. (This is a normal subgroup of SL2 (Fq ).) Define P SL2 (Fq ) to be the quotient SL2 (Fq )/Z. This is called the projective special linear group of degree 2 over Fq . Definition 243 (2nd version) Define the projective line P1 (Fq ) of the finite field Fq to be the q + 1 values of the formal ratio x/y, where x, y run over all elements of Fq . If y = 0 then we denote this formal value by ∞, so P1 (Fq ) = Fq ∪ {∞}. If q = p is a prime then we denote P1 (Fp ) = {∞, 0, 1, ..., p − 1}. Define P SL2 (Fq ) to be the group of all Mobius transformations on the projective line f (x) = (ax + b)/(cx + d), x ∈ P1 (Fq ), where ad − bc = 1 and a, b, c, d ∈ Fq (We define f (∞) = a/c.) Mobius transformations are bijections from the projective line to itself, so we may interpret each Mobius transformation as an element of SX , where X = P1 (Fq ) (and therefore also of Sn , where n = |P1 (Fq )| = q + 1). Example 244 Let p = 11 and let f (x) = −1/x. Then x f (x) ∞ 0 0 ∞ 1 10 2 5 3 7 4 8 5 2 6 9 7 3 8 4 9 6 10 1

254

CHAPTER 13. CROSSING THE RUBICON

Therefore, as a permutation, f = (∞, 0)(1, 10)(2, 5)(3, 7)(4, 8)(6, 9). The following facts are known about the projective special linear group: Theorem 245 If q > 3 then P SL2 (Fq ) is a simple group. Moreover, for all prime powers q, |P SL2 (Fq )| = (q 2 − 1)q/gcd(2, q − 1). (Recall a simple group was defined in Definition 162.) This theorem is over 100 years old. It is proven, for example, in [R]. Theorem 246 Choose a k ∈ Fq such that < k >= F× q . Let f1 (x) = x + 1,

f2 (x) = k · x,

f3 (x) = −1/x.

Then P SL2 (Fq ) is generated by f1 , f2 , and f3 . In particular, the action of P SL2 (Fq ) on the projective line X = P1 (Fq ) yields an injective homomorphism P SL2 (Fq ) → SX . Basically, this is proven in [R] as well.

13.3

Galois’ last dream

Supposedly, the night before he died in a duel, Galois wrote a letter to a friend stating the following remarkable theorem: Theorem 247 (Galois) Assume p > 11. Then P SL2 (Fp ) has no embedding into a symmetric group Sn with n ≤ p. The following isomorphisms (for q ≤ 11) are known: P SL2 (Fq ) ∼ =

   A4 , q = 3,  

A5 , q = 5, A6 , q = 9.

If p = 7 or p = 11 then explicit embeddings of P SL2 (Fp ) into A8 (p = 7), A12 (p = 11) are known (see [CS], ch 10, or [K] for an excellent discussion of this).

13.4. THE M12 GENERATION

13.4

255

The M12 generation

One of the most amazing aspects about M12 is its close relationship with other ”interesting” groups. Definition 248 Define the permutation f4 of the set P1 (Fp ) = {∞, 0, 1, ..., p− 1}, for 3 ≤ p ≤ 11, as follows

f4 =

        

1, p = 3, (1, 2)(3, 4), p = 5, (1, 2)(3, 6), p = 7, (2, 10)(3, 4)(5, 9)(6, 7), p = 11.

We have run across the group S6 before, when studying the symmetries of the icosahedron. We have also seen that S6 is rather an interesting group because it is the only non-abelian symmetric group Sn which has an outer automorphism. One rather connection between M12 and S6 is given by the following Theorem 249 (a) If p = 5 then S6 =< f1 , f2 , f3 , f4 >. (b) If p = 11 then M12 =< f1 , f2 , f3 , f4 >. We shall see another interpretation of M12 below using coding theory! Definition 250 Let (

δ(x) =

x3 /9, x ∈ (F23 )2 − 0, 9x3 , x ∈ P1 (F23 ) − (F23 )2 .

This is an element of SX , where X = P1 (F23 ). Define the Mathieu group M24 by < f, δ | f ∈ P SL2 (F23 ) > . This is a permutation group in SX , where X = P1 (F23 ). By the way, (a) (F23 )2 = {0, 1, 2, 3, 4, 6, 8, 9, 12, 13, 16, 18}, (b) M12 = 8 · 9 · 10 · 11 · 12 = 95040, (c) |M24 | = 244823040.

256

CHAPTER 13. CROSSING THE RUBICON

13.5

Coding the Golay way

Codes are used in everyday life, from ISBN numbers on books to barcodes on food products to music CDs to satellite transmissions. There are many types of codes, some more efficient than others, some with better error correcting ability than others, some more practical than others, and so on. We shall concern ourselves only with aspects which are related to (in one way or another) permutation puzzles. Definition 251 A q-ary code is a subset C of a finite dimensional vector space V over the finite field Fq . A code word is an element of C. The number of coordinates (i.e., the dimension of V) is called the length of the code word. If q = 2 then the code is called binary (instead of 2-ary) and if q = 3 then the code is called ternary (instead of 3-ary). Example 252 V = Fnq = Fq × ... × Fq (n times) is a code. Definition 253 Let M onn (Fq ) denote the group of all n × n matrices which have exactly one non-zero entry from Fq per row and per column. An element of M onn (Fq ) is called a monomial matrix. Exercise 13.5.1 (a) Show M onn (Fq ) is a group. (b) Show |M onn (Fq )| = (q − 1)n · n!. Definition 254 The set of all A ∈ M onn (Fq ) such that A ∗ C = C (i.e., are the same code) is called the automorphism group of C, denoted Aut(C). We shall see some examples below. Definition 255 If w is a code word in Fnq then number of non-zero coordinates of w is called the weight of w, denoted wt(w). A cyclic code is a code which has the property that whenever (c0 , c1 , ..., cn−1 ) is a code word then so is (cn−1 , c0 , ..., cn−2 ). If (c0 , c1 , ..., cn−1 ) is a code word in a cyclic code V then we call g(x) = c0 + c1 x + ... + cn−1 xn−1 a generator polynomial for C .

13.5. CODING THE GOLAY WAY

257

Example 256 The code with elements (1, 1, 0, 1, 0, 0, 0), (0, 1, 1, 0, 1, 0, 0), (0, 0, 1, 1, 0, 1, 0), (0, 0, 0, 1, 1, 0, 1) is a binary cyclic code of length 7 and weight 3. It has generator polynomial g(x) = 1 + x + x3 . (This code is called a ”Hamming code” and has many interesting properties which, to describe, would take us too far afield. The interested reader is refered to [CS], ch. 3.) Definition 257 Let n be a positive integer relatively prime to q and let alpha be a primitive n-th root of unity. Each generator polynomial g of a cyclic code C of length n has a factorization of the form g(x) = (x − αk1 )...(x − αkr ), where {k1 , ..., kr } ⊂ {0, ..., n − 1}. The numbers αki , 1 ≤ i ≤ r, are called the zeros of the code C. They do not depend on the choice of g. Definition 258 Let p and n be distinct primes and assume that p is a square mod n. The quadratic residue code of length n over Fp is the cyclic code whose generator polynomial has zeros {αk | k is a square mod n}. The binary Golay code GC23 is the quadratic residue code of length 23 over F2 . The binary Golay code GC24 is the code of length 24 over F2 obtained by appending onto GC23 a zero-sum check digit. The ternary Golay code GC11 is the quadratic residue code of length 11 over F3 . The ternary Golay code GC12 is the code of length 12 over F3 obtained by appending onto GC11 a zero-sum check digit. The following result illustrate how the Matheiu groups arise in coding theory. Theorem 259 (a) There is a normal subgroup N of Aut(GC12 ) of order 2 such that Aut(GC12 )/N is isomorphic to M12 . (b) Aut(GC24 ) = M24 .

258

CHAPTER 13. CROSSING THE RUBICON

Since the Mathieu groups are so large, this theorem above indicates that the Golay codes GC12 and GC24 have a lot of symmetry. It is a basic rule of thumb in mathematics that whenever you find something displaying a lot of symmetry then it will quite often have other interesting properties. With this philosophy spurring us on, let us turn to some of the other properties of these codes. Lemma 260 Any two code words in GC24 differ by 8 bits. The code GC24 detects 4 errors (per 24 bits) and corrects 3 errors. When you compare that with the correcting ability of bar-codes or ISBN codes (which have a check-digit), GC24 is much better. Lemma 261 If w is a code word in GC24 then wt(w) is either 0, 8, 12, 16, 24. Definition 262 The code words of weight 12 in GC24 are called dodecads. We may identify a code word w = (c0 , c1 , ..., c23 ) with the set of indices i of the non-zero coordinates ci 6= 0. Theorem 263 M12 is the stabilizer in M24 of a dodecad, regarded as a set of indices.

13.6

M12 is crossing the rubicon

Let f1 , f2 , ..., f12 denote the basic moves (2π/5 degree turns of a ”pentagon” about a vertex) of the Rubik isocahedron, regarded as elements of SV , where V denotes the set of 12 vertices of the Rubik isocahedron (”rubicon”). The following remarkable result is due to John Conway [CS]. Theorem 264 M12 =< x ∗ y −1 | x, y ∈ {f1 , ..., f12 } >. In other words, the Mathieu group M12 is generated by the twist-untwist moves of the Rubik isocahedron. If we call a ”twist-untwist” move of the form x ∗ y −1 (with x, y as in the theorem above) a cross move then (with apologies to Caeser) the theorem above says that M12 is generated by the crosses of the rubicon. Question: Label the vertices of the megaminx 1, 2, ..., 20 in any way you like. Let C denote the cross subgroup of S20 generated by the cross moves of the megaminx.

13.7. AN ASIDE: A PAIR OF CUTE FACTS

259

(1) Is C simple? (2) Does C act (via the cross moves above) 5-transitively on the set of vertices of the megaminx? The answer to (1) is, according to GAP [Gap], ”yes”. In fact, GAP implies the following Theorem 265 The cross subgroup of S20 is the alternating group on 20 letters: C ∼ = A20 . From this and Theorem 139, it follows that the answer to (2) is ”yes” as well. In fact, C acts 18-transitively on the set of vertices of the megaminx. The subgroup S of the megaminx group generated by all elements of the form x ∗ y −1 , where x, y correspond to faces which have no intesection is called the slice group of the megaminx. Problem (Longridge): Determine S. An analogous problem exists for the Rubicon.

13.7

An aside: A pair of cute facts

It’s hard to resist stating some more interesting facts about the Mathieu groups.

13.7.1

Hadamard matrices

Let A = (aij )1≤i,j≤n denote a real n×n matrix. The following question seems quite natural in a course in advanced vector calulus or real analysis: Question: What is the maximum value of | det(A)|, where the entries of A range over all real numbers |aij | ≤ 1? From vector calculus we know that the absolute value of the determinant of a real square matrix equals the volume of the parallelpiped spanned by the row (or column) vectors of the matrix. The volume of a parallelpiped with sides of a fixed length depends on the angles the row vectors make with each other. This volume is maximized when the row vectors are mutually orthogonal, i.e., when the parallelpiped is a cube in Rn . Suppose now that the row vectors of A are all orthogonal. The row vectors of A, |aij | ≤ 1, are longest√ when each aij = ±1, which implies that the length of each row vector is √ n. Suppose, in addition, that the row vectors of A are all of length n. Such a matrix is called a Hadamard matrix of order n. Then

260

CHAPTER 13. CROSSING THE RUBICON

√ √ | det(A)| = ( n)n = nn/2 , since the cube has n sides of length n. Now, if A is any matrix as in the above question then we must have | det(A)| ≤ nn/2 . This inequality is called Hadamard’s inequality. What is shocking at first (at least to me) is that, there does not always exist a Hadamard matrix. For example, there is a 2 × 2 Hadamard matrix but not a 3 × 3 one. What is perhaps even more suprising is that, in spite of the fact that the above question (which is unsolved) arose from an analytic perspective, Hadamard matrices are related more to coding theory, number theory, and combinatorics [vLW]! Example 266 Let              A :=            

1 −1 −1 −1 −1 −1 −1 −1 −1 −1 −1 −1

1 1 −1 1 −1 −1 −1 1 1 1 −1 1

1 1 1 −1 1 −1 −1 −1 1 1 1 −1

1 1 1 1 −1 1 −1 −1 −1 1 1 1

1 1 −1 1 1 −1 1 −1 −1 −1 1 1

1 1 1 −1 1 1 −1 1 −1 −1 −1 1

1 1 1 1 −1 1 1 −1 1 −1 −1 −1

1 −1 1 1 1 −1 1 1 −1 1 −1 −1

1 −1 −1 1 1 1 −1 1 1 −1 1 −1

1 −1 −1 −1 1 1 1 −1 1 1 −1 1

1 1 −1 −1 −1 1 1 1 −1 1 1 −1

1 −1 1 −1 −1 −1 1 1 1 −1 1 1

                        

This is a Hadamard matric of order 12. Exercise 13.7.1 Show that (a) if you swap two rows or columns of a Hadamard matrix, you will get another Hadamard matrix, (b) if you multiply any row or column of a Hadamard matrix by −1, you will get another Hadamard matrix, (c) if you multiply any Hadamard matrix on the left by a signed permutation matrix (that is, a matrix with exactly one ±1 per row and column) then you will get another Hadamard matrix, (d) if you multiply any Hadamard matrix on the left by a signed permutation matrix (that is, a matrix with exactly one ±1 per row and column) then you will get another Hadamard matrix.

13.7. AN ASIDE: A PAIR OF CUTE FACTS

261

Exercise 13.7.2 Let A, B be two Hadamard matrices of order n. Call A, B left equivalent if there is an n × n signed permutation matrix P such that A = P B. Show that this defines an equivalence relation on the set of all Hadamard matrices of order n. Exercise 13.7.3 Let A be a Hadamard matrix of order n. Let Aut(A) denote the set of all n×n signed permutation matrices Q such that A is left equivalent to AQ. Show that Aut(A), called the automorphism group of A, is a group under matrix multiplication. The following result is yet another indication of the unique role of these Mathieu groups in mathematics: Theorem 267 (M. Hall, Assmus-Mattson [AM]) Let A be the Hadamard matrix of order 12 in the above example. Then Aut(A) ∼ = M12 .

13.7.2

5-transitivity

The following result exemplifies once more the unique role of these Mathieu groups in group theory: Theorem 268 If G is a subgroup of SX for some finite set X and if G acts 5-transitively on X then exactly one of the following must be true: (a) G ∼ = Sn , for some n > 4, (b) G ∼ = Am , for some m > 6, ∼ (c) G = M12 , (d) G ∼ = M24 . Furthermore, each of the groups in (a)-(d) acts 5-transitively on some finite set. For a proof of this, see [CS] and [R], ch 9.

262

CHAPTER 13. CROSSING THE RUBICON

Chapter 14 Appendix: Some solution strategies ”The emphasis on mathematical methods seems to be shifted more towards combinatorics and set theory - and away from the algorithm of diiferential equations which dominates mathematical physics.” J. von Neumann and O. Morganstern, THEORY OF GAMES AND ECOMONIC BEHAVIOR, 1944 This chapter includes some strategies for solving the 3x3 Rubik’s cube, the 4x4 Rubik’s cube, the masterball, the equator puzzle, the skewb, and the pyraminx. First (this is a mathematics course, after all!) we discuss some of the mathematical ideas behind the computer algorithms used to study the Rubik’s cube:

14.1

The subgroup method

One approach to solve the Rubik’s cube using a computer has been to construct a certain sequence of subgroups Gn = {1} < Gn−1 < ... < G1 < G0 = G, where G =< R, L, F, B, U, D > is the Rubik’s cube group, which allows the following strategy to be implemented: 263

264

CHAPTER 14. APPENDIX: SOME SOLUTION STRATEGIES

• represent a given position of the Rubik’s cube by an element g0 ∈ G, • determine a complete set of coset representatives of Gk+1 /Gk : k Gk+1 /Gk = ∪ri=1 gk+1,i Gk+1 ,

some rk > 1, ∀0 ≤ k < n

(note mn−1 = 1, gn,1 = 1), • (step 1) if g0 ∈ g1,i G1 (where i ∈ {1, ..., n1 }) then let g1 = g1,i and g10 = g1−1 g0 (note g10 ∈ G1 ), • (inductive step) if gk0 ∈ Gk has been defined and if gk0 ∈ gk+1,j Gk −1 0 0 (where j ∈ {1, ..., n1 }) then let gk+1 = gk+1,j and gk+1 = gk+1 gk (note 0 gk+1 ∈ Gk+1 ), −1 −1 • putting all these together, we obtain 1 = gn−1 gn−1 gn−2 ...g1−1 g0 , so

g0 = g1 g2 ...gn−1 gn . The hope is to be able to choose the sequence of subgroups Gi in such a way that the coset representatives are short, relatively simple moves on the Rubik’s cube so that the ”solution” g0 = g1 g2 ...gn−1 gn is not too long.

14.1.1

Example: the corner-edge method

We now present an example - a fairly unsophisticated one but you will get the idea. Let G1 denote the subgroup which does not permute any corners, let G2 denote the subgroup which does not permute any corners or edges, let G3 denote the subgroup which does not permute any corners or edges and does not reorient any corners, and let G4 = {1}: G4 = {1} < G3 < G2 < G1 < G0 = G. This choice of subgroups crudely models the ”corner-edge method” (see the appendix) due to Singmaster [Si]. The idea is simple. 1. Represent a given position of the Rubik’s cube by an element g0 ∈ G.

14.1. THE SUBGROUP METHOD

265

2. Let g1 denote the move which moves all the corners into the correct positions (i.e., permutes them into the solved position and possibly twists them), so g1−1 g0 ∈ G1 . Let g10 = g1−1 g0 . 3. Let g2 denote the move which moves all the edges into the correct positions (i.e., permutes them into the solved position and possibly reorients corners and edges) and leaves all other pieces unpermuted, so g2−1 g10 ∈ G2 . Let g20 = g2−1 g10 . 4. Let g3 denote the move which ”solves” all the corners (i.e., twists them all into the correct orientation and may flip some edges) but does not permute any pieces, so g3−1 g20 ∈ G3 . Let g30 = g3−1 g20 . 5. Let g4 denote the move which ”solves” all the edges (i.e., flips them all into the correct orientation) and leaves all other facets along. 6. The ”solution” is g0 = g1 g2 g3 g4 .

14.1.2

Example: Thistlethwaite’s method

Morwen Thistlethwaite (a knot-theorist now at the Univ. of Tennessee) developed one of the best subgroup methods for solving the cube [FS]. He takes G1 =< R, L, F, B, U 2 , D2 >, G2 =< R, L, F 2 , B 2 , U 2 , D2 >, G3 =< R2 , L2 , F 2 , B 2 , U 2 , D2 >, G4 = {1}. G2 is isomorphic to the ”Rubik’s 3 × 3 × 2-domino” group. It’s order is (8!)2 · 12, according to [FS], §7.6. G3 is the ”squares” group. It’s order is 213 · 34 , according to [FS], §7.6. He has shown (using a computer to help with some of the work) that • there is a complete set of coset representatives {g1,i | 1 ≤ i ≤ n1 } of G/G1 such that each g1,i is at most 7 moves long (and n1 = 2048), • there is a complete set of coset representatives {g2,i | 1 ≤ i ≤ n2 } of G1 /G2 such that each g2,i is at most 13 moves long (and n2 = 1082565), • there is a complete set of coset representatives {g3,i | 1 ≤ i ≤ n3 } of G2 /G3 such that each g3,i is at most 15 moves long (and n3 = 29400),

266

CHAPTER 14. APPENDIX: SOME SOLUTION STRATEGIES

• there is a complete set of coset representatives {g4,i | 1 ≤ i ≤ n4 } of G3 /G4 such that each g4,i is at most 17 moves long (and n4 = 663552).

Therefore, the Rubik’s cube can be solved in at most 7 + 13 + 15 + 17 = 52 moves. More recent improvement’s on this method have gotten this number down to fourty-something I think (see [Lo] for details and recent updates).

14.2

3 × 3 Rubik’s cube

Consider the group G =< R, L, U, D, F, B > of moves of the Rubik’s cube. The size of the group generated by these permutations is 43252003274489856000 ∼ = 19 4.3 × 10 .

14.2.1

Strategy for solving the cube

Let xy = y −1 ∗ x ∗ y denote conjugation and [x, y] = x ∗ y ∗ x−1 ∗ y −1 denote the commutator, for x, y group elements. Let MR denote clockwise (with respect to right side) quarter turn of the middle slice parallel to the right side. The layer method solution strategy is composed of 3 stages: Stage 1: Solve the top face and top edges. Stage 2: Solve the middle edges (and bottom edges as best as possible). Stage 3: Solve the bottom corners (and bottom edges if necessary). The corner-edge method solution strategy is composed of 2 stages: Stage 1: Solve and then orient the corners. (The move U ∗ F ∗ [R, U ]3 ∗ F −1 permutes (ubr,ufl)(uf,ul,ub,ur).) Stage 2: Solve the and then orient the edges. ”clean” edge and corner moves:

14.3. 4 × 4 RUBIK’S CUBE MR2 ∗ U −1 ∗ MR−1 ∗ U 2 ∗ MR ∗ U −1 ∗ MR2 (MR ∗ U )3 ∗ U ∗ (MR−1 ∗ U )3 ∗ U (R2 U 2 )3 (MR ∗ U )4 ((D2 )R ∗ (U 2 )B )2 [R, U ]3 2 2 2 F L U (F 2 L2 )3 U 2 L2 F 2 (D2 R2 D2 (F 2 R2 )2 U )2 (MR2 U MR2 U 2 )2 −1 [DR , U ]

267 edge 3-cycle (uf,ul,ur) flips the top edges uf, ub permutes (uf,ub)(fr,br) flips ub,ul and flips df,db ufr+, bld++ permutes (ufr,dfr)(ubr,ubl) permutes (uf,ub)(ur,ul) permutes (ufl,ubr)(dfr,dbl) permutes (ufl,ubr)(ufr,ubl) corner 3-cycle (brd,urb,ulb)

These moves were compiled with help from the books [Si], [B], [Sn], and [Gap].

14.2.2

Catalog of 3 × 3 Rubik’s ”supercube” moves

The supercube is the Rubik’s cube with each center facet marked with a short line through it and an adjoining edge. * (MR2 ∗U −1 ∗MR−1 ∗U 2 ∗MR ∗U −1 ∗MR2 )2 is the top edge 3-cycle (uf,ur,ul), * (R−1 ∗ D2 ∗ R ∗ B −1 ∗ U 2 ∗ B)2 twists the ufr corner clockwise and the bld corner counterclockwise (and does not twist any centers). * MR−1 ∗ MD−1 ∗ MR ∗ U ∗ MR−1 ∗ MD ∗ MR−1 ∗ U is the center twist u+, r(for these last three moves, see [Si], [Sn])

14.3

4 × 4 Rubik’s cube

The solution strategy is composed of 3 stages: Stage 1: Solve the corners. For this moves for the 3 × 3 Rubik’s cube. Stage 2: ”Pair” the edges so that the neighboring facets on neighboring middle edges have the same color. For this the following ”clean edge moves” are useful: • flipedge: L22 ∗ D12 ∗ U2 ∗ F13 ∗ U23 ∗ F1 ∗ D12 ∗ L22 ∗ L1 ∗ U1 ∗ L31 ∗ U23 ∗ L1 ∗ U13 ∗ L31 (due to J. Adams [A] who calls it ”move 8”). This flips and swaps the two middle edge facets on the U F boundary. It affects some centers, but no other edges or corners.

268

CHAPTER 14. APPENDIX: SOME SOLUTION STRATEGIES

• upedgeswap: R2 ∗ B12 ∗ D12 ∗ B13 ∗ R23 ∗ B1 ∗ D12 ∗ B13 ∗ R2 ∗ B13 ∗ R22 (due to Thai [T], who calls it an ”11 gram”). This move affects some centers but no corners and only 4 edge facets. It swaps and flips the right-most UF edge cubie with the left-most (with respect to the B face) UB edge subcube, sending the U facet of the right-most UF edge subcube to the B facet of the left-most UB edge subcube. • 3-cycle (R2 ∗ U1 )3 is a 3 cycle on the edges: (R2 uf, B2 ur, B2 ul). This doesn’t affect any other edges and leaves all corners fixed. Stage 3: Solve the edges. For this the ”clean edge moves” for the 3 × 3 Rubik’s cube. Stage 4: If necessary, apply the flipedge move above, Stage 5: Solve the centers. For this, use the following ”clean center move”: center3cycle = R1−1 ∗ F2 ∗ R2−1 ∗ F2−1 ∗ R1 ∗ F2 ∗ R2 ∗ F2−1 (also called ”move 9”, due to J. Adams). This move is a 3-cycle on centers facets, affecting no edges, no corners, and no other center facets. It is the 3-cycle (15 19 18) in the above notation. Some similar clean center moves: center1 = B12 ∗ R23 ∗ F2 ∗ R2 ∗ B12 ∗ R23 ∗ F23 ∗ R2 , center2 = R22 ∗ B12 ∗ R23 ∗ F2 ∗ R2 ∗ B12 ∗ R23 ∗ F23 ∗ R23 These aren’t really necessary since the the center3cycle can always be applied after a suitable set-up move (i.e., in combination with a suitable conjugation). The following move is occasionally useful: centerswap = (R22 ∗ U22 )4 This affects only 6 center facets (on the front and back faces) and no others. It is the product of two 3-cycles: (15 34 23)(27 14 22) in the above notation. These moves were compiled with help from the books [A] and [T].

14.4. RAINBOW MASTERBALL

14.4

269

Rainbow masterball

The solution strategy Step 1: The idea is to first get all the middle bands aligned first, so you get ball corresponding to a matrix of the form * 1 1 *

* 2 2 *

* 3 3 *

* 4 4 *

* 5 5 *

* 6 6 *

* 7 7 *

* 8 8 *

Here, ∗ denotes any color. We have labeled the colors on the masterball as 1, 2, ..., 8 in order of occurrence. We describe a method, which I call ”fishing”, for achieving this. (Mathematically, this amounts to performing some carefully choosen commutators.) Without too much trouble you can always assume that we have one column aligned. You may need to flip or rotate the ball a little bit to do this. Call this aligned column ”column 1” and call the color in column 1, ”color 1”. We want to get the middle two entries in column 2 aligned. Call the color in the (2,3)-entry ”color 2”. We want to get color 2 in the (2,2)-entry. The remaining large color 2 tile is what we will ”fish” for. Hold the ball in front of you in such a way that column 2 is slightly to the left of center and column 3 is slightly to the right of center. There are 4 facets in the right upper middle band, 4 facets in the left upper middle band, 4 facets in the right lower middle band and 4 facets in the left lower middle band. A flip about the center on the right half (i.e., perform f2 ) exchanges these. We may assume that color 2 is on one of the four facets in the right lower middle band. (If it isn’t you need to apply f2 first). Now perform r2−1 ∗ f2−1 ∗ r2 ∗ f2 : first perform r2−1 (this is ”baiting the hook”), then f2−1 (”putting the hook in the water”), then r2 (”setting the hook”), and finally f2 (”reeling in the hook”). You may or may not have color 2 in the (2,2) place like you want but the color 1 stripe is intact. If necessary, try again. After at most 4 tries you’ll be successful. Step 2: Repeat this ”crab fishing” strategy to get color 2 in the (1,2) position (using r1−1 ∗ f2−1 ∗ r1 ∗ f2 in place of r2−1 ∗ f2−1 ∗ r2 ∗ f2 ). Now, by turning the ball over if necessary, repeat this idea to get color 2 in the (4,2) position. Now you have two ”aligned” stripes on your ball - color 1 in column 1 and color 2 in column 2. We say, in this case, that columns 1 and 2 have been ”solved”.

270

CHAPTER 14. APPENDIX: SOME SOLUTION STRATEGIES Step 3: Repeat this for columns 3 and 4. Step 4: Use the moves in the ”catalog” below to finish the puzzle.

14.4.1

A catalog of rainbow moves

Column moves We number the columns as 1,...,8. We will use a signed cycle notation to denote an action of a move on the columns of the masterball. Example 269 A move which switches the 1st and 3rd column but flips both of them over will be denoted by (1 3)− . A move which sends the 4th column to the 6th column, the 6th column to the 5th column, and switches the 2nd and 3rd column but flips both of them over will be denoted by (2 3)− (6 5 4). move f1 f2 f3 f4 f5 f6 f7 f8 f1 ∗ f2 ∗ f1 f1 ∗ f2 ∗ f1 ∗ f2 f1 ∗ f3 ∗ f1 f2 ∗ f3 ∗ f2 f1 ∗ f4 ∗ f1 f1 ∗ f5 ∗ f1 f1 ∗ f8 ∗ f1 f8 ∗ f1 ∗ f8 f2 ∗ f1 ∗ f2 f3 ∗ f1 ∗ f3 f8 ∗ f1 ∗ f2

cycle (1, 4) (2, 3) (2, 5) (3, 4) (3, 6) (4, 5) (4, 7) (5, 6) (5, 8) (6, 7) (1, 6) (7, 8) (2, 7) (1, 8) (3, 8) (1, 2) (1, 2) (3, 5) (5, 4, 3, 2, 1) (1, 5)(2, 6) (2, 3) (6, 5, 4) (1, 7)(5, 6) (5, 8) (6, 7) (2, 8)(3, 4) (1, 8) (4, 3, 2) (1, 3)(4, 5) (1, 3)(4, 8) (1, 4) (2, 3, 8, 5)

Finally, (f1 ∗ f2 ∗ f3 ∗ f4 )2 ∗ r1 ∗ r2 ∗ r3 ∗ r4 swaps the 7,8 columns and leaves all the others fixed but flipped over.

14.5. EQUATOR PUZZLE

271

Some products of 2-cycles on the facets These are all based on an idea of Andrew Southern. The polar2swap and equator2swap were obtained by trying variations of some of Andrew’s moves on a MAPLE implementation of the masterball [J]. We number the facets in the i-th column, north-to-south, as i1, i2, i3, i4 (where i = 1, 2, ..., 8). move x = r1 ∗ f4 ∗ r1−1 ∗ r4 ∗ f4 ∗ r4−1 x ∗ r14 ∗ x ∗ r44 f1 ∗ r1 ∗ f4 ∗ r1−1 ∗ r4 ∗ f4 ∗ r4−1 ∗ f1 polar2swap36 polar2swap18 equator2swap36 equator2swap18

(41, (41, (14, (11, (61, (12, (62,

cycle 84)(44, 81)(44, 84)(11, 14)(31, 64)(11, 13)(32, 63)(12,

81) 84) 81) 61) 81) 62) 82)

where polar2swap36 = f1 ∗r3−1 ∗r4−1 ∗f1 ∗f2 ∗r1 ∗r4−1 ∗f2 ∗r44 ∗f2 ∗r1−1 ∗r4 ∗f2 ∗r44 ∗f1 ∗r3 ∗r4 ∗f1 (moreover, if you replace r3 by r2 both times in this move you get the same effect), polar2swap18 = f1 ∗r3−1 ∗r4−1 ∗f3 ∗f4 ∗r1 ∗r4−1 ∗f4 ∗r44 ∗f4 ∗r1−1 ∗r4 ∗f4 ∗r44 ∗f3 ∗r3 ∗r4 ∗f1 equator2swap36 = f1 ∗r4−1 ∗r3−1 ∗f1 ∗f2 ∗r2 ∗r3−1 ∗f2 ∗r34 ∗f2 ∗r2−1 ∗r3 ∗f2 ∗r34 ∗f1 ∗r4 ∗r3 ∗f1 equator2swap18 = f1 ∗r4−1 ∗r3−1 ∗f3 ∗f4 ∗r2 ∗r3−1 ∗f4 ∗r34 ∗f4 ∗r2−1 ∗r3 ∗f4 ∗r34 ∗f3 ∗r4 ∗r3 ∗f1 For further details on the rainbow puzzle, see [JS], [J].

14.5

Equator puzzle

Solution strategy, in brief: First, ignore the orientation of the pieces. Just try to get the pieces in their correct position. One of the most remarkable properties of this puzzle is that GAP [Gap] is, in practice, very efficient at solving this part of the solution. (This is remarkable in view of the fact that GAP is not very good

272

CHAPTER 14. APPENDIX: SOME SOLUTION STRATEGIES

at solving the corresponding problem for the Rubik’s cube, for example, so there is no reason to expect it to be good at solving this problem.) Second, once the pieces are in the correct position, they must be correctly oriented by a catalog of ”node” moves designed for that purpose. Some ”node” moves are included below. Some notation: if x, y, and z are moves, let [x, y, z] = x−1 ∗ y −1 ∗ z −1 ∗ x ∗ y ∗ z.

We call a move of the form [r13k , r23m , r33n ] a node move since it only affects the nodes (where the circles intersect). The table below records where each node goes as well as its effect on the orientation. For example, [r1−3 , r2−3 , r3−3 ] will swap the NP and SP, while rotating the NP by 90 degrees in a counter clockwise direction. Moreover, it will fix the position of the piece labeled as 20 but will rotate it by 90 degrees clockwise. The position entry in a block is the position the piece moves to. The angle entry, if any is the angle the piece gets rotated by. No angle entry means, of course, that the piece is not rotated. No position entry means that the piece was not moved (but may have been rotated). If a move has no effect on the position or the angle then we fill in the block with a ”-”. In the following table, NP, SP have been denoted by 1,7, resp., for brevity. move,piece m123 m132 m231 m213 m312 m321 A m3212 B n123 D

1 7,90ccw 7,180 7,180 90cw 90ccw 7,90ccw 180 1,90ccw -

7 1,90cw 1,180 1,180 90ccw 90cw 1,90cw 180 7,90cw -

4 10,180 90ccw 90cw 10,90cw 10,90cw 10,180 180 180 10 90cw

10 4,180 90cw 90ccw 4,90ccw 4,90ccw 4,180 180 180 4 90ccw

15 90ccw 20,90cw 20,90cw 20 20 90cw 180 20,90ccw 90cw

20 90cw 15,90ccw 15,90ccw 15 15 90ccw 180 15,90cw 90ccw

14.6. THE SKEWB where

273

m123 = [r1−3 , r2−3 , r3−3 ], m132 = [r1−3 , r3−3 , r2−3 ], m231 = [r2−3 , r3−3 , r1−3 ], m213 = [r2−3 , r1−3 , r3−3 ], A = m123 ∗ m132 ∗ m213, B = C ∗ m3212 ∗ C −1 , C = (1 4)(7 10) = r13 , n123 = [r1−3 , r2−3 , r33 ], D = [r1−3 , r2−3 , r33 ].

14.6

The skewb

14.6.1

Strategy

The goal here is to collect enough moves to support the following solution strategy: fix the centers and solve the corners using ”clean corner moves” (i.e., moves which do not effect the centers). The basic moves are twists by 120 degrees clockwise about each of the six corners FRU,FLU,BRU,BLU,BDR,BDL,DFR,DFL.

14.6.2

A catalog of skewb moves

Thanks to J. Montague and G. Gomes for comments and corrections for the descriptions below. 1. F RU ∗ BLU ∗ F RU −1 is order 3. 2. [F RU ∗ F LU ]3 twists 6 corners clockwise by 120 degrees. The 2 corners not twisted are those opposite the FRU, FLU corners: the BDR, BDL corners. The centers are all fixed. (F RU ∗ F LU )3 = (F LU ∗ F RU )3 rotates all the corners except for the bd corners. It does not permute any facets. 3. The move, [F RU ∗ BLU ]5 , fixes all the centers and the 2 ”opposite” corners: DFL, BDR. It twists the 3 corners FLU, BLU, and FRU. On the remaining 3 corners, it acts as the permutation (DF R, BRU, BDL). 4. FRU = BDL (actually, they are only equal up to a rotation of the entire cube). In general, a corner move is equal to the opposite corner move up to a rotation of the entire cube.

274

CHAPTER 14. APPENDIX: SOME SOLUTION STRATEGIES

5. (F RU ∗ F LU )3 ∗ (BDL ∗ BDR)6 rotates all the corners except for the bu and the df corners. The uf corners are rotated clockwise and the bd corners counterclockwise. It does not permute any facets. (F RU ∗ F LU )6 ∗ (BDL ∗ BDR)3 is the same move, but rotates in the opposite direction. 6. (F RU ∗ F LU )6 ∗ (BDL ∗ BDR)6 rotates the corners as follows: • the uf corners counterclockwise, • the db corners counterclockwise, • the df corners clockwise, • the ub corners clockwise. It does not permute any facets. 7. Let bottomspin = (F RU ∗ F LU )6 ∗ (BDL ∗ BDR)3 ∗ (DF R ∗ DF L)3 . This move rotates the 4 bottom corners (the df corners clockwise and the db counterclockwise). It does not permute any facets. 8. (BRU ∗ F LU )9 is a 5 cycle on the center facets (F, R, B, U, L). It fixes the bottom and does not affect any corners. (BLU ∗ F RU )9 is a 5 cycle on the center facets. It fixes the bottom and does not affect any corners. 9. (BLU ∗ F RU )9 ∗ (BRU ∗ F LU )9 is a product of 2 transpositions on the center facets, swapping front/back and up/right. It fixes the bottom and does not affect any corners. 10. Let bottomspin = (F RU ∗ F LU )6 ∗ (BDL ∗ BDR)3 ∗ (DF R ∗ DF L)3 , and let U denote the clockwise (wrt up face) rotation of entire cube by 90 degrees. Then bottomspin∗U ∗bottomspin rotates but does not swap 2 corners (the DFR and BDL) and does not affect any other corners or faces.

14.7

The pyraminx

Assume that the tetrahedron is lying on a flat surface in front of you, with the triangle base pointing away from you. The corners are denoted L (left), R (right), U (up), and B (back).

14.8. THE MEGAMINX

275

Basic Moves: Let • L denote the 120 degree clockwise rotation of the 2-level subtetrahedron containing the left corner, • R denote the 120 degree clockwise rotation of the 2-level subtetrahedron containing the right corner, • U denote the 120 degree clockwise rotation of the 2-level subtetrahedron containing the up corner, • B denote the 120 degree clockwise rotation of the 2-level subtetrahedron containing the back corner. First, get the ”center” facets solved, then twist the corner tips to solve them and the center facets. Finally, to solve the edge facets, use the following moves (given in [EK]): • [R, U −1 ] is a 3-cycle of edge pieces on the URL face, • [R, U −1 ] ∗ [R−1 , L] is a flip of two edges (UR edge and UL edge) on the URL face.

14.8

The megaminx

The strategy here is the same as for the 3x3 Rubik’s cube: • place the corners correctly first (ignoring correct corner orientation), • place the edges correctly first (ignoring correct edge orientation), • twist the corners if necesary, • flip the edges if necessary. Moves useful for carrying out these steps are included in the following catalog.

276

CHAPTER 14. APPENDIX: SOME SOLUTION STRATEGIES

14.8.1

Catalog of moves

First, some notation. We label the faces f1 , f2 , ..., f6 on top and label the bottom faces f7 , f8 , ..., f12 as in chapter 4. The same notation is used to indicate the move of the megaminx given by rotating that face of the megaminx by 72 degrees clockwise. • f1−1 ∗ f2−1 ∗ f1 ∗ f2 ∗ f1−1 ∗ f2−1 ∗ f1 ∗ f2 ∗ f1−1 ∗ f2−1 ∗ f1 ∗ f2 = [f1 , f2 ]3 swaps the f1 .f3 and the f2 .f6 corners: (f1 .f2 .f3 , f1 .f3 .f4 )(f1 .f2 .f6 , f2 .f6 .f7 ) • m = f3 ∗f6−1 ∗f4 ∗f2−1 ∗f5 ∗f3−1 ∗f6 ∗f4−1 ∗f2 ∗f5−1 ; f6 ∗f1 ∗m6 ∗f1−1 ∗f6−1 swaps 2 pairs of corners on the f1 face: (f1 .f2 .f6 , f1 .f3 .f4 )(f1 .f2 .f3 , f1 .f5 .f6 ) • f1 ∗ f6 ∗ f1−1 ∗ f2 ∗ f1 ∗ f6−1 ∗ f1−1 ∗ f2−1 3 cycle on corners and 3-cycle on edges (f1 .f2 .f6 , f1 1.f7 .f6 , f2 .f6 .f7 )(f1 .f2 , f6 .f7 , f2 .f6 ) • M 2 = f6 ∗ f2 ∗ f1 ∗ f2−1 ∗ f1−1 ∗ f6−1 ∗ f3−1 ∗ f1−1 ∗ f2−1 ∗ f1 ∗ f2 ∗ f3 (Mark Longridge) edge 3-cycle (f1 .f2 , f2 .f3 , f2 .f6 ) • (f6−1 ∗ f2−1 ∗ f3−1 ∗ f6 ∗ f2 ∗ f3 )6 - triple corner twister, ccw twists of f1 .f2 .f3 , f1 .f2 .f6 , f1 .f5 .f6 (

• M 3 = f3 − 2) ∗ f62 ∗ f2 ∗ f1−1 ∗ f6 ∗ f1 ∗ f32 ∗ f62 ∗ f1 ∗ f6−2 ∗ f3−2 ∗ f1−1 ∗ f6−1 ∗ f1 ∗ f2−1 ∗ f6−2 ∗ f32 ∗ f1−1 (Mark Longridge) edge 2-flip of f1 .f2 , f1 .f6 • M 3a = f6−1 ∗ f2−1 ∗ f1 ∗ f3−1 ∗ f1−1 ∗ f3 ∗ f2 ∗ f6 ∗ f3 ∗ f2 ∗ f1−1 ∗ f6 ∗ f1 ∗ f6−1 ∗ f2−1 ∗ f3−1 (Mark Longridge) edge 2-flip of f1 .f2 , f1 .f3 For further details, see the internet sites [J] or [Lo].

Bibliography [A] J. Adams, HOW TO SOLVE RUBIK’S REVENGE, Dial Press, NY, 1982 [Are] Andrew Arensburger, Square 1 www page http://www.cfar.umd.edu/~arensb/Square1/ [Ar] M. Artin, ALGEBRA, Prentice-Hall, 1991 [AM] E. Assmus, Jr. and H. Mattson, ”On the auotmorphism groups of Paley-Hadamard matrices”, in COMBINATORIAL MATHEMATICS AND ITS APPLICATIONS, ed. R. Bose, T. Dowling, Univ of North Carolina Press, Chapel Hill, 1969 [Ba] J. Baez, ”Some thoughts on the number 6”, internet newsgroup sci.math article, posted May 22, 1992, http://math.ucr.edu/home/baez/README.html [B] C. Bandelow, INSIDE RUBIK’S CUBE AND BEYOND, Birkhauser Boston, 1980 [BCG] Berlekamp, J. Conway, R. Guy, WINNING WAYS, II, Academic Press, [BH] R. Banerji and D. Hecker, ”The slice group in Rubik’s cube”, Math. Mag. 58(1985)211-218 [Bu] G. Butler, FUNDAMENTAL ALGORITHMS FOR PERMUTATION GROUPS, Springer-Verlag, Lecture Notes in Computer Science, 559, 1991 277

278

BIBLIOGRAPHY

[Car] R. Carter, SIMPLE GROUPS OF LIE TYPE, Wiley, 1972 [CD] M. Conrady and M. Dunivan, ”The Cross Group of the Rubik’s Cube”, SM485C project, April,1997 [CS] J. Conway and N. Sloane, SPHERE PACKINGS, LATTICES, AND GROUPS, Springer-Verlag, 1993 [CFS] G. Cooperman, L. Finkelstein and N. Sarawagi, ”Applications of Cayley graphs”, in APPLIED ALGEBRA ..., Springer-Verlag, Lecture Notes in Computer Science, 508, 1990 [C] J. Crossley, et al, WHAT IS MATHEMATICAL LOGIC?, Dover, 1972 [CL] ftp archives of the ”cube-lovers” list at ftp://ftp.ai.mit.edu/pub/cubelovers/ [CG] S. Curran and J. Gallian, ”Hamiltonian cycles and paths in Cayley graphs and diagraphs - survey”, Discrete Math. 156(1996)1-18 [DL] M. Dunbar and A. Luers, ”The Group Structure of the 2x2 Rubik’s Cube”, SM485 term paper, Fall 1996 [EK] J. Ewing and C. Kosniowski, PUZZLE IT OUT, CUBES, GROUPS, AND PUZZLES, Cambridge Univ Press, 1982 [FH] W. Fulton and J. Harris, REPRESENTATION THEORY, Springerverlag, 1991 [FS] A. Frey and D. Singmaster, HANDBOOK OF CUBIK MATH, Enslow Pub., 1982 [G] A. Gaglione, AN INTRODUCTION TO GROUP THEORY, NRL, 1992 [Gap] Martin Sch¨onert et al, GAP MANUAL, Lehrstuhl D f¨ ur Mathematik, RWTH Aachen [GJ] M. Garey and D. Johnson, COMPUTERS AND INTRACTIBILITY, W. H. Freeman, New York, 1979 [Gar1] M. Gardner, ”Combinatorial card problems” in TIME TRAVEL AND OTHER MATHEMATICAL BEWILDERMENTS, W. H. Freeman, New York, 1988

BIBLIOGRAPHY

279

[Gar2] M. Gardner, MY BEST MATHEMATICAL AND LOGIC PUZZLES, Dover, New York, 1994 [GT] K. Gold, E. Turner, ”Rubik’s group”, Amer. Math. Monthly, 92(1985)617-629 [GM] G. Gomes and J. Montague, ”The skewb group”, SM485C project, April,1997 [HR] G. Hardy and S. Ramanujan, ”Asymptotic formulae in combinatory analysis”, Proc. London Math. Soc. 17(1918)75-115 [H] D. Hofstadler, METAMATHEMATICAL THEMAS, Basics Books, 1985 (Mostly a collection of Scientific American columns he wrote; the articles referred to here were also published in Scientific American, March 1981, July 1982) [Hum] J. Humphreys, REFLECTION GROUPS GROUPS, Cambridge Univ Press, 1990 [I]

AND

COXETER

J. Isbell, ”The Gordon game of a finite group”, Amer. Math. Monthly 99(1992)567-569

[J] D. Joyner, ”Rainbow masterball page”, internet www page http://www.nadn.navy.mil/MathDept/~wdj/rainbow.html [JM] D. Joyner and J. McShea, ”The homology group of the square 1 puzzle”, preprint [JN] D. Joyner and G. Nakos, LINEAR ALGEBRA AND APPLICATIONS, to be published (PW+S, 1998?) [JS] D. Joyner and A. Southern, ”The masterball puzle”, preprint [Ki] A. Kirillov, ELEMENTS OF THE THEORY OF REPRESENTATIONS, Springer-Verlag, 1976 [K] B. Kostant, ”The graph of the truncated icosahedron and the last letter of Galois”, Notices of the A.M.S. 42(1995)959-968

280

BIBLIOGRAPHY

[L] M. E. Larsen, ”Rubik’s revenge: the group theoretical solution”, Amer. Math. Monthly, 92(1985)381-390 [Lo] M. Longridge, ”God’s Algorithm Calculations for Rubik’s Cube, Rubik’s Subgroups, and Related Puzzles”, internet www page http://web.idirect.com/~cubeman/ [Lu] A. Luers, ”The group structure of the pyraminx and the dodecahedral faces of M12 ”, USNA Honors thesis, 1997 (Advisor W. D. Joyner) http://web.usna.navy.mil/~wdj/m_12.htm [Ma] G. Mackey, UNITARY GROUP REPRESENTATIONS IN PHYSICS, PROBABILITY, AND NUMBER THEORY, Math Lecture Notes Series, Benjamin/Cummins, 1978 [MKS] W. Magnus, A. Karrus and D. Solitar, Combinatorial Group Theory, 2nd ed, Dover, 1976 [Mc] J. McShea, ”The 14-15 Puzzle, and Why It Can’t Be Solved”, SM485 term paper, Fall 1996 [M] R. E. Moritz, MEMORABILIA MATHEMATICA, MacMillan Co, NY, 1914 [NST] P. Neumann, G. Stoy and E. Thompson, GROUPS AND GEOMETRY, Oxford Univ. Press, 1994 [Rob] S. Robinson, ”The Mathematics of Bell Ringing”, capstone paper (Advisor W. D. Joyner) [R] J. J. Rotman, AN INTRODUCTION TO THE THEORY OF GROUPS, 4th ed, Springer-Verlag, Grad Texts in Math 148, 1995 [Ru] E. Rubik, et al, RUBIK’S CUBIC COMPENDIUM, Oxford Univ Press, 1987 [S] R. Schmalz, OUT OF THE MOUTHS OF MATHEMATICIANS, Math. Assoc. Amer., 1993

BIBLIOGRAPHY

281

[Se] J.-P. Serre, LINEAR REPRESENTATIONS OF FINITE GROUPS, Springer-Verlag, 1977 [Ser] J.-P. Serre, TREES, Springer-Verlag, 1980 [Si] D. Singmaster, NOTES ON RUBIK’S MAGIC CUBE, Enslow, 1981 [Sn] R. Snyder, GET CUBED [Sn2] R. Snyder, TURN TO SQUARE 1, 1993 [St] R. Stoll, SET THEORY AND LOGIC, Dover, 1963 [TW] A. D. Thomas and G. V. Wood, GROUP TABLES, Shiva Publishing Ltd, Kent, UK, 1980 [T] Thai, “The winning solution to Rubik’s Revenge”, Banbury Books, 1982 [vLW] J. van Lint and R. M. Wilson, A COURSE IN COMBINATORICS, Cambridge Univ. Press, 1992 [Wa] H. B. Walters, CHURCH BELLS OF ENGLAND, Oxford Univ Press, 1912 [Wh] White, Arthur, ”Fabian Stedman: The First Group Theorist?”, American Mathematical Monthly, Nov. 1996, pp771-8. [W] R. M. Wilson, “Graph puzzles, homotopy, and the alternating group”, J. of Combin. Theory, 16 (1974)86-96